Pith. sign in

REVIEW 3 major objections 5 minor 8 cited by

Quantization-Aware Imitation-Learning for Resource-Efficient Robotic Control

T0 review · 3 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper claims that a two-part quantization-aware imitation learning loss recovers the accuracy of full-precision robotic policies at 4-bit precision, preserving near-baseline success while cutting memory several-fold and speeding up…

desk verdict A practical quantized-IL recipe with credible hardware numbers, but the accuracy claims need error bars and the distillation setup has an unaddressed covariate-shift gap. read the letter →

arxiv 2412.01034 v1 pith:M6NO472C submitted 2024-12-02 cs.RO cs.CVcs.LG

Seongmin Park , Hyungmin Kim , Wonseok Jeon , Juyoung Yang , Byeongwook Jeon , Yoonseon Oh , Jungwook Choi This is my paper · ORCID
classification cs.ROcs.CVcs.LG
keywords quantization-awaretrainingimitationlearningbehaviorcloningvision-language-actionmodelsautonomousdrivingrobotmanipulationlow-bitquantizationedgedeployment
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Large imitation-learning policies such as vision-language-action models are too slow and memory-hungry for onboard robots, but naive low-bit quantization degrades action quality. The paper proposes quantization-aware imitation learning (QAIL) combined with quantization-robust behavior cloning (QBC): fine-tune the quantized policy on expert demonstrations while also matching the action distribution of the full-precision teacher. On OpenVLA with INT4 weights, the method reaches a 73.1 percent average success rate on LIBERO versus 74.0 percent for the Bfloat16 baseline, and on the CILRS driving model at W4A4 it approaches full-precision success on NoCrash-dense while cutting infractions. The paper reports up to 2.5x speedup and 2.5x energy savings on an edge GPU for manipulation, and up to 3.7x speedup and 3.1x energy savings on a low-end GPU for driving. If the claim holds, aggressive 4-bit quantization becomes a practical path for deploying large robotic policies on resource-limited hardware without sacrificing decision quality.

What carries the argument

The machinery is a sum of two losses: the QAIL loss $\mathcal{L}_{\text{IL}} = -\mathbb{E}_{(s_t,a_t)\sim\mathcal{D}_{\text{QAIL}}} \log \pi^q_\theta(a_t|s_t)$, trained on expert demonstrations merged with state-action pairs collected from the full-precision policy, and the QBC loss $\mathcal{L}_{\text{QBC}} = \mathbb{E}_{s_t\sim\pi^q_\theta} D(\pi^q_\theta(a_t|s_t), \pi_{\text{FP}}(a_t|s_t))$, which pushes the quantized policy's action distribution toward the full-precision teacher's. A weighted version increases the QBC term on states deemed important by a perturbation-based saliency score over the image. The two losses do complementary work: QAIL anchors the policy to demonstrator actions, while QBC prevents the distribution from being distorted by quantization errors, and their combination is what the paper credits for recovering baseline-level performance.

What would settle it

Run the QAIL+QBC trained quantized policy in closed loop, record the states it actually visits, and compare them against the states in the training set $\mathcal{D}_{\text{QAIL}}$; if the distribution gap is large, or if removing the full-precision rollouts from $\mathcal{D}_{\text{QAIL}}$ collapses success to QAIL-only levels, then the QBC benefit depends on a training-distribution assumption that may not hold at deployment.

Watch

Extended reading notes

Core claim

The central claim is that a policy's robustness to low-bit quantization errors can be trained into it through imitation learning, and that the first successful recovery of full-precision performance for quantized IL-based policies is achieved by combining two losses: QAIL, which maximizes the likelihood of actions from expert demonstrations plus full-precision policy rollouts, and QBC, which minimizes the divergence between the quantized policy's action distribution and the full-precision policy's distribution over states. Together these losses keep the quantized policy faithful to the teacher's overall behavior while retaining the high-quality actions present in expert data. Concretely, OpenVLA with INT4 weights reaches 73.1 percent average success on LIBERO compared to 74.0 percent for the Bfloat16 baseline, and CILRS at W4A4 reaches near full-precision success rates on NoCrash-dense with reduced pedestrian collisions and red-light infractions. A saliency-weighted variant of QBC improves long-horizon manipulation tasks, and the appendix extends the same QBC idea to quantized reinforcement learning policies.

Load-bearing premise

The QBC loss assumes that the states the quantized policy will meet during deployment are well-covered by the full-precision policy rollouts and expert demonstrations used during training, and the paper does not directly validate that coverage.

Editorial extensions

If this is right

  • A 7.6B-parameter vision-language-action model can run at INT4 weights with a roughly one-percentage-point drop in LIBERO success versus Bfloat16, while using about one quarter of the memory.
  • A W4A4 self-driving model can reach near full-precision success and driving scores on NoCrash-dense, with pedestrian collisions reduced to near full-precision levels.
  • On the measured hardware, deploying these quantized policies yields up to 2.5x speedup and 2.5x energy savings for the manipulation model, and up to 3.7x speedup and 3.1x energy savings for the driving model, with 1.7x speedup and 1.3x energy savings for an 8-bit CPU deployment.
  • Applying saliency-weighted QBC improves long-horizon task success, raising LIBERO-Long from 47.8 percent with plain QBC to 50.4 percent.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The QBC term is effectively a policy-distillation objective from the full-precision teacher, so the same idea could likely transfer to other sequential decision-making settings where a trusted full-precision policy exists and a quantized student must be deployed under latency or memory constraints.
  • The training data for QAIL is collected from full-precision rollouts rather than from the quantized policy's own state distribution, leaving a covariate-shift gap; collecting rollouts from the evolving quantized policy, in the style of DAgger, would be a natural test of whether closing that gap pushes the INT4 results even closer to baseline.
  • The reported speedups and energy savings are tied to the specific inference stacks and kernels used on the Jetson and 2080Ti platforms, and may not transfer to other hardware without re-tuning.
  • The attention-divergence analysis suggests QAIL+QBC restores the quantized policy's focus to task-relevant regions; a follow-up could test whether the same alignment holds under distribution shift, such as novel object layouts or weather conditions.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes quantization-aware imitation learning (QAIL) combined with quantization-robust behavior cloning (QBC) to fine-tune quantized policy models for robot manipulation, autonomous driving, and continuous control. The core idea is to fine-tune a quantized policy with an imitation-learning loss on expert plus full-precision rollouts while also minimizing a distributional discrepancy against the full-precision policy. Experiments with OpenVLA on LIBERO, CILRS on NoCrash-dense, and D4RL/DeepMind Control Suite report near-full-precision success rates with 4-bit weights or weights-and-activations, together with 2.5–3.7× speedups and comparable energy savings on edge hardware such as the Jetson AGX Orin and RTX 2080 Ti. The paper claims this is the first approach to recover and deploy the performance of quantized IL-based policies successfully.

Significance. If the empirical claims hold, the paper demonstrates a practically important result: aggressive 4-bit quantization can be applied to large imitation-learning policies (including a 7.6B-parameter VLA model) with only a small accuracy penalty and substantial deployment gains on resource-limited hardware. The strengths of the paper include real hardware measurements with detailed latency/energy breakdowns, evaluation across three distinct policy classes, and comparison to prior quantization-aware RL methods (LPPD, QuaRL). The attention and action-distribution analyses provide useful qualitative insight even if they are not fully independent evidence. However, the central accuracy claim currently rests on single-seed comparisons with no error bars, and the QBC loss is defined on a state distribution that the algorithm does not actually sample from; both issues need to be resolved before the headline conclusions can be accepted.

major comments (3)
  1. [§4.2, Eq. (3) vs. Algorithm 1] The QBC loss is defined as L_QBC = E_{s_t ~ π^q_θ} D(π^q_θ(a_t|s_t), π_FP(a_t|s_t)), i.e., an expectation over states visited by the quantized policy. Algorithm 1 instead populates D_QAIL = D_FP ∪ D_E, where D_FP is collected by rolling out the full-precision policy π_FP, and both L_QAIL and λL_QBC are evaluated on those states. The training state distribution therefore differs from the deployment distribution of the quantized policy. This is load-bearing because the paper's claim of 'robustness throughout the sequence' (Sec. 4.2) relies on QBC regularizing the quantized policy's own covariate-shifted state distribution; with the implemented objective, the distillation signal is only provided on states the quantized policy may never visit. Please clarify which distribution is actually used in the experiments. If QBC is indeed computed off-policy, provide an ablation that measures the mismatch, or a theoretical argument that the off-policy objective still controls compounding errors.
  2. [§5.2, Tables 3–5] No error bars, confidence intervals, or numbers of seeds are reported anywhere. The headline result of Table 3 is a 73.1% versus 74.0% average success rate (a 0.9-point gap) on LIBERO, where each suite is evaluated over 500 episodes; with binomial sampling noise, this gap is within one standard error, and the task-level swings in Table 4 (e.g., task8: 70 vs. 82, task6: 88 vs. 94) are also consistent with noise. Similarly, Table 5 shows per-condition differences of up to 10 percentage points between QAIL+QBC and the FP baseline. Without repeated seeds or confidence intervals, the central claim that QAIL+QBC 'preserves' full-precision accuracy cannot be statistically assessed. Please report variance over multiple fine-tuning runs or at least binomial confidence intervals for the success rates.
  3. [§4.3, Table 2] The attention-divergence analysis is partly self-confirming: QBC, as defined in Eq. (3), explicitly minimizes a discrepancy between the quantized and full-precision action distributions, so reporting that QAIL+QBC yields lower AttDiv than AWQ is expected from the construction and does not provide independent evidence that the method improves task-relevant reasoning. Please either reframe Table 2 as a sanity check of the optimization objective, or use a metric that is not directly optimized (e.g., task success conditioned on attention fixation, or a behavioral measure such as object-touch counts).
minor comments (5)
  1. [§1 and §6] The phrase 'first to recover and deploy the performance of quantized IL-based policies successfully' is stronger than what the evidence supports, given the absence of statistical error bars and the existence of prior quantization-aware policy methods (LPPD, QuaRL) that the paper itself compares against; please temper the claim.
  2. [Algorithm 1] The pseudocode states 'for each state-action pair (s_t, a_t) in D_QAIL' and computes L_total = L_IL + λL_QBC, but Eq. (3) requires full action distributions rather than a single action; the pseudocode should specify how D(·,·) is computed in practice (e.g., average L2 over sampled actions, or a closed-form distance for Gaussian policies).
  3. [§4.2, Eq. (7)] The wQBC formulation introduces three hyperparameters (β, threshold T, and the 20% / 10% selection rule) but Table 1 reports only a single LIBERO-Long comparison with no sensitivity analysis. A short ablation over β or T would help establish that the reported 2.6-point improvement is not an artifact of a single threshold choice.
  4. [§A.1.2 and §A.4.1] The main text cites the NVIDIA tegrastats tool [34] for energy measurement, while the appendix says the jetson-stats library is used; please reconcile which tool actually produced the reported energy numbers, since this affects reproducibility.
  5. [Throughout] There are several typos and minor formatting issues, including 'Autonomus Driving' in Section A.1.1, 'training proceeds wih a learning rate' in Section A.1.2, and the rendering of Eq. (6) where '1/I × J' appears garbled; a careful proofread is needed.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: central claims rest on external benchmark comparisons, not on definitions or self-citations.

full rationale

The paper's central accuracy claims (Tables 3, 5, and 7) are evaluated on external benchmarks (LIBERO, NoCrash, D4RL) against FP/Bfloat16 baselines and prior methods (AWQ, RTN, LPPD, QuaRL). The reported success rates are empirical outcomes, not quantities fitted from the loss definitions. QBC's loss in Eq. (3) minimizes a distributional distance between the quantized and FP policies' action distributions, but the performance gains in Tables 3-5 are not equal to that loss by construction; they are measured task-success rates. The attention-divergence analysis in Table 2 uses Eq. (8), which compares saliency maps, not the action-distribution discrepancy minimized by Eq. (3), so the lower AttDiv for QAIL+QBC is an empirical observation rather than a tautology. The only self-citation that overlaps with the authors ([38]) appears as a related-work pointer to structured pruning and is not load-bearing for the paper's claims. The mismatch between Eq. (3), which samples states from the quantized policy, and Algorithm 1, which trains on D_QAIL = D_FP ∪ D_E, is a real covariate-shift/validity concern, but it is not circularity: the predictions are not defined in terms of the inputs, and the central benchmark comparisons remain independent of that mismatch. No prediction in the paper reduces by definition to its training objective or to a self-citation chain.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The paper introduces no new physical entities or conserved quantities. The central claim rests on five main assumptions: the quality of expert demonstrations, the validity of STE gradients, coverage of the deployment state distribution by training rollouts, the error-accumulation premise, and the reliability of perturbation saliency. The free parameters are hyperparameters and learned quantization scales that affect the reported results but are not the product of a derived theory.

free parameters (5)
  • QBC weight lambda = 1
    Set to 1 for all task domains in Appendix A.1.2; balances the QAIL and QBC losses and directly affects the final success rates.
  • wQBC scaling beta = 2
    Applied only to top-20% salient states in LIBERO-Long via Eq. (7); chosen as a hyperparameter greater than 1 without a reported ablation sweep.
  • Saliency threshold T = top 20%
    Differentiates important states for wQBC; computed from 10% of the fine-tuning dataset in Section 4.2, which is a hand-selected proportion.
  • LSQ step-size parameters = learned per tensor
    Used for CILRS and D4RL weight/activation quantization; these scales are trained parameters from LSQ and are essential to the reported W4A4 results.
  • QLoRA rank r = 32
    Rank of the LoRA adapters used to fine-tune OpenVLA in Section 5.1; a hyperparameter that controls trainable capacity (110M parameters).
assumptions (5)
  • domain assumption Behavior cloning objective (Eq. 1) assumes expert demonstrations contain the action distribution that should be imitated.
    The QAIL loss in Eq. (2) maximizes the likelihood of expert and FP-policy actions; if the expert data are suboptimal, the fine-tuned quantized policy inherits that suboptimality.
  • standard math The straight-through estimator provides usable gradients through the quantization operation.
    Algorithm 1 replaces the quantization gradient with STE during backpropagation, a standard approximation that is not derived or validated in the paper.
  • domain assumption States encountered by the quantized policy during deployment are sufficiently covered by FP-policy and expert rollouts used to build D_QAIL.
    Eq. (3) samples from pi^q_theta, but Algorithm 1 collects states from pi_fp and expert data; the paper does not address the resulting covariate shift.
  • domain assumption Quantization errors accumulate over action sequences and can be counteracted by aligning the quantized policy's action distribution to the FP policy.
    This is the motivating premise of QBC in Section 3.2 and Section 4.2, and it underpins the choice of the QBC loss.
  • domain assumption Perturbation-based saliency (Eq. 5) faithfully identifies states where accurate action cloning matters most.
    wQBC up-weights the loss by S_pi; if saliency does not correlate with decision-critical states, the weighting can distort the learned policy.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quantization-Aware Imitation-Learning for Resource-Efficient Robotic Control." pith.science (2026). https://pith.science/paper/M6NO472C

@misc{pith2026241201034,
  author       = {Pith},
  title        = {Pith review of: Quantization-Aware Imitation-Learning for Resource-Efficient Robotic Control},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/M6NO472C}},
  note         = {Machine review of arXiv:2412.01034}
}
read the original abstract

Deep neural network (DNN)-based policy models like vision-language-action (VLA) models are transformative in automating complex decision-making across applications by interpreting multi-modal data. However, scaling these models greatly increases computational costs, which presents challenges in fields like robot manipulation and autonomous driving that require quick, accurate responses. To address the need for deployment on resource-limited hardware, we propose a new quantization framework for IL-based policy models that fine-tunes parameters to enhance robustness against low-bit precision errors during training, thereby maintaining efficiency and reliability under constrained conditions. Our evaluations with representative robot manipulation for 4-bit weight-quantization on a real edge GPU demonstrate that our framework achieves up to 2.5x speedup and 2.5x energy savings while preserving accuracy. For 4-bit weight and activation quantized self-driving models, the framework achieves up to 3.7x speedup and 3.1x energy saving on a low-end GPU. These results highlight the practical potential of deploying IL-based policy models on resource-constrained devices.

Figures

Figures reproduced from arXiv: 2412.01034 by the authors.

Figure 1
Figure 1. (a) Differences in robot action between INT4 quantization and Bfloat16 in [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Comparison of the structure, number of parameters of DNN-based policy models. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Overview of QAIL+QBC closely replicates expert behavior. This is often achieved through behavior cloning, which optimizes a supervised learning objective to maximize the likelihood of expert state-action pairs in the dataset. The objective loss function is: L IL(θ) = − 1 |DE| X (s,a)∈DE log πθ(a|s). (1) Minimizing this loss enables the learned policy πθ to imitate the expert by predicting expert actions a for given … view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Action accuracy Comparison. To address QAIL’s challenge, we propose quantization￾robust behavior cloning (QBC), leveraging the FP pol￾icy alongside demonstration data to minimize quantization errors by aligning the action distributions of the quantized and FP policies.…
Figure 5
Figure 5. Figure 5: Comparison of attention visualization for tasks successfully completed on the LIBERO-Spatial benchmark. Additional examples [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Comparison of action distributions at a specific state [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: LIBERO benchmark. In our experiments, we used datasets that were specif￾ically modified for compatibility with the OpenVLA [24] framework, which included enhancements such as high￾resolution image processing, image rotation, and the exclu￾sion of unsuccessful demonstra…
Figure 8
Figure 8. Figure 8: Comparison of attention visualization for tasks successfully completed on the LIBERO-Spatial benchmark. [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]
Figure 9
Figure 9. Figure 9: Overview of QARL+QBC with PPO 3. Integration of QBC: To enhance π q ’s performance fur￾ther and ensure stability in complex scenarios, we integrate the QBC as introduced in equation 3 of Sec. 4.2. QBC aids in aligning the decision-making of the quantized policy with th…
Figure 10
Figure 10. Figure 10: Hardware execution timeline comparison of 16-bit, 8-bit, and 4-bit weight datatypes for [PITH_FULL_IMAGE:figures/full_fig_p016_10.png]
Figure 11
Figure 11. Figure 11: Timeline comparison of OpenVLA for 16-bit, 8-bit, and 4-bit datatypes over a total duration of 1,000 ms. V represents the ViT+MLP projector, P denotes the prefill operation, and D refers to the decode operation of the backbone LLM (Llama-2-7b) [PITH_FULL_IMAGE:figure…
Figure 12
Figure 12. Figure 12: Latency breakdown of OpenVLA for 16-bit, 8-bit, and 4-bit weights within a single step, including ViT+MLP, Llama-2- 7b (Prefill), and Llama-2-7b (Decode) stages. a timeline comparison for each weight type based on the ac￾tual time ratio, illustrating the proportion of…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 8 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Bridging the Semantic-Action Gap in Visual Token Pruning for Efficient VLA Inference

    cs.CV 2025-11 conditional novelty 6.0 of 10

    Token pruning for VLA models should use both prefill semantic attention and temporally smoothed action-decode attention; VLA-Pruner does this and preserves manipulation performance at high prune ratios.

  2. SQAP-VLA: A Synergistic Quantization-Aware Pruning Framework for High-Performance Vision-Language-Action Models

    cs.CV 2025-09 conditional novelty 6.0 of 10

    A training-free framework jointly quantizes a VLA model to 4 bits and prunes visual tokens, recovering or exceeding full-precision success rates at 1.93x speedup.

  3. SpecPrune-VLA: Accelerating Vision-Language-Action Models via Action-Aware Self-Speculative Pruning

    cs.CV 2025-09 conditional novelty 5.0 of 10

    SpecPrune-VLA accelerates OpenVLA-style robot policies by pruning visual tokens using global attention from prior frames, early-layer speculation, and an action-speed-aware controller, achieving ~1.5x speedup on LIBER...

  4. Source Component Shift Adaptation via Offline Decomposition and Online Mixing Approach

    cs.LG 2025-08 unverdicted novelty 5.0 of 10

    The claimed result is that source-component-shift adaptation splits cleanly into offline component learning via EM and online mixing-weight updates, cutting cumulative test loss by up to 67.4%.

  5. CEED-VLA: Consistency Vision-Language-Action Model with Early-Exit Decoding

    cs.RO 2025-06 conditional novelty 5.0 of 10

    CEED-VLA uses consistency distillation plus a fixed early-exit point to accelerate VLA action decoding 2.0x to 4.1x with negligible success-rate loss across CALVIN, LIBERO, and real-robot trials.

  6. Leveraging OS-Level Primitives for Robotic Action Management

    cs.OS 2025-08 conditional novelty 4.0 of 10

    Applying OS-style exception handling, context caching, and replay to robotic action slices raises success rates 7x to 24x and cuts execution steps up to 74% for repetitive manipulation tasks, without retraining the VLA model.

  7. RLRC: Reinforcement Learning-based Recovery for Compressed Vision-Language-Action Models

    cs.RO 2025-06 conditional novelty 4.0 of 10

    RLRC combines 90% structured pruning, supervised fine-tuning, PPO reinforcement learning, and optional 4-bit quantization to recover OpenVLA's success rate on ManiSkill while cutting memory up to 8x and boosting throu...

  8. Think Twice, Act Once: Token-Aware Compression and Action Reuse for Efficient Inference in Vision-Language-Action Models

    cs.CV 2025-05 conditional novelty 4.0 of 10

    FlashVLA, a training-free plug-in, reuses stable actions and prunes visual tokens to cut VLA model inference FLOPs by 55.7% and latency by 36% with only a 0.7% success-rate drop on LIBERO.

Reference graph

Works this paper leans on

56 extracted references · 23 canonical work pages · cited by 8 Pith papers

  1. [1]

    Flamingo: a visual language model for few-shot learning

    Jean-Baptiste Alayrac, Jeff Donahue, Pauline Luc, Antoine Miech, Iain Barr, Yana Hasson, Karel Lenc, Arthur Men- sch, Katherine Millican, Malcolm Reynolds, et al. Flamingo: a visual language model for few-shot learning. Advances in neural information processing systems, 35:23716–23736,

  2. [2]

    Post train- ing 4-bit quantization of convolutional networks for rapid- deployment

    Ron Banner, Yury Nahshan, and Daniel Soudry. Post train- ing 4-bit quantization of convolutional networks for rapid- deployment. Advances in Neural Information Processing Systems, 32, 2019. 2

    2019

  3. [3]

    Distributed dis- tributional deterministic policy gradients

    Gabriel Barth-Maron, Matthew W Hoffman, David Bud- den, Will Dabney, Dan Horgan, Dhruva Tb, Alistair Mul- dal, Nicolas Heess, and Timothy Lillicrap. Distributed dis- tributional deterministic policy gradients. arXiv preprint arXiv:1804.08617, 2018. 15

    arXiv 2018

  4. [4]

    Rt-1: Robotics transformer for real-world control at scale

    Anthony Brohan, Noah Brown, Justice Carbajal, Yevgen Chebotar, Joseph Dabis, Chelsea Finn, Keerthana Gopalakr- ishnan, Karol Hausman, Alex Herzog, Jasmine Hsu, et al. Rt-1: Robotics transformer for real-world control at scale. arXiv preprint arXiv:2212.06817, 2022. 1

    arXiv 2022

  5. [5]

    Rt-2: Vision-language-action models transfer web knowledge to robotic control

    Anthony Brohan, Noah Brown, Justice Carbajal, Yevgen Chebotar, Xi Chen, Krzysztof Choromanski, Tianli Ding, Danny Driess, Avinava Dubey, Chelsea Finn, et al. Rt-2: Vision-language-action models transfer web knowledge to robotic control. arXiv preprint arXiv:2307.15818, 2023. 1, 3

    arXiv 2023

  6. [6]

    {TVM}: An automated {End- to-End} optimizing compiler for deep learning

    Tianqi Chen, Thierry Moreau, Ziheng Jiang, Lianmin Zheng, Eddie Yan, Haichen Shen, Meghan Cowan, Leyuan Wang, Yuwei Hu, Luis Ceze, et al. {TVM}: An automated {End- to-End} optimizing compiler for deep learning. In 13th USENIX Symposium on Operating Systems Design and Im- plementation (OSDI 18), pages 578–594, 2018. 8

    2018

  7. [7]

    Gonzalez, Ion Stoica, and Eric P

    Wei-Lin Chiang, Zhuohan Li, Zi Lin, Ying Sheng, Zhang- hao Wu, Hao Zhang, Lianmin Zheng, Siyuan Zhuang, Yong- hao Zhuang, Joseph E. Gonzalez, Ion Stoica, and Eric P. Xing. Vicuna: An open-source chatbot impressing gpt-4 with 90%* chatgpt quality, 2023. 1

    2023

  8. [8]

    Pact: Parameterized clipping activa- tion for quantized neural networks

    Jungwook Choi, Zhuo Wang, Swagath Venkataramani, Pierce I-Jen Chuang, Vijayalakshmi Srinivasan, and Kailash Gopalakrishnan. Pact: Parameterized clipping activa- tion for quantized neural networks. arXiv preprint arXiv:1805.06085, 2018. 1, 2, 3, 4

    arXiv 2018

Show all 56 references
  1. [9]

    Xception: Deep learning with depthwise separable convolutions

    Franc ¸ois Chollet. Xception: Deep learning with depthwise separable convolutions. In Proceedings of the IEEE con- ference on computer vision and pattern recognition , pages 1251–1258, 2017. 2

    2017

  2. [10]

    Exploring the limitations of behavior cloning for autonomous driving

    Felipe Codevilla, Eder Santana, Antonio M L ´opez, and Adrien Gaidon. Exploring the limitations of behavior cloning for autonomous driving. In Proceedings of the IEEE/CVF international conference on computer vision , pages 9329–9338, 2019. 2, 6, 12

    2019

  3. [11]

    Qlora: Efficient finetuning of quantized llms

    Tim Dettmers, Artidoro Pagnoni, Ari Holtzman, and Luke Zettlemoyer. Qlora: Efficient finetuning of quantized llms. Advances in Neural Information Processing Systems , 36,

  4. [12]

    Carla: An open urban driv- ing simulator

    Alexey Dosovitskiy, German Ros, Felipe Codevilla, Anto- nio Lopez, and Vladlen Koltun. Carla: An open urban driv- ing simulator. In Conference on robot learning, pages 1–16. PMLR, 2017. 6, 12

    2017

  5. [13]

    Palm- e: An embodied multimodal language model

    Danny Driess, Fei Xia, Mehdi SM Sajjadi, Corey Lynch, Aakanksha Chowdhery, Brian Ichter, Ayzaan Wahid, Jonathan Tompson, Quan Vuong, Tianhe Yu, et al. Palm- e: An embodied multimodal language model. arXiv preprint arXiv:2303.03378, 2023. 1

    2023 arXiv

  6. [14]

    Learned step size quantization

    Steven K Esser, Jeffrey L McKinstry, Deepika Bablani, Rathinakumar Appuswamy, and Dharmendra S Modha. Learned step size quantization. In International Conference on Learning Representations, 2020. 1, 2, 3, 4, 6, 12

    2020

  7. [15]

    Quarl: Quantization for fast and environ- mentally sustainable reinforcement learning

    Aleksandra Faust, Gabe Barth-Maron, Maximilian Lam, Sharad Chitlangia, Srivatsan Krishnan, Vijay Janapa Reddi, and Zishen Wan. Quarl: Quantization for fast and environ- mentally sustainable reinforcement learning. Transactions on Machine Learning Research (TMLR) 2022 , 2022. 3, 14, 15

    2022

  8. [16]

    OPTQ: Accurate quantization for generative pre- trained transformers

    Elias Frantar, Saleh Ashkboos, Torsten Hoefler, and Dan Alistarh. OPTQ: Accurate quantization for generative pre- trained transformers. In The Eleventh International Confer- ence on Learning Representations, 2023. 1, 2

    2023

  9. [17]

    D4rl: Datasets for deep data-driven rein- forcement learning

    Justin Fu, Aviral Kumar, Ofir Nachum, George Tucker, and Sergey Levine. D4rl: Datasets for deep data-driven rein- forcement learning. arXiv preprint arXiv:2004.07219, 2020. 3, 6, 8

    2004 arXiv

  10. [18]

    Visualizing and understanding atari agents

    Samuel Greydanus, Anurag Koul, Jonathan Dodge, and Alan Fern. Visualizing and understanding atari agents. In Inter- national conference on machine learning, pages 1792–1801. PMLR, 2018. 5

    2018

  11. [19]

    Apple in- telligence foundation language models

    Tom Gunter, Zirui Wang, Chong Wang, Ruoming Pang, Andy Narayanan, Aonan Zhang, Bowen Zhang, Chen Chen, Chung-Cheng Chiu, David Qiu, et al. Apple in- telligence foundation language models. arXiv preprint arXiv:2407.21075, 2024. 3

    2024

  12. [20]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceed- ings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016. 2

    2016

  13. [21]

    1.1 computing’s energy problem (and what we can do about it)

    Mark Horowitz. 1.1 computing’s energy problem (and what we can do about it). In 2014 IEEE international solid-state circuits conference digest of technical papers (ISSCC), pages 10–14. IEEE, 2014. 2

    2014

  14. [22]

    Fixed-point feedfor- ward deep neural network design using weights +1, 0, and -1

    Kyuyeon Hwang and Wonyong Sung. Fixed-point feedfor- ward deep neural network design using weights +1, 0, and -1. In Signal Processing Systems (SiPS), 2014 IEEE Workshop on, pages 1–6. IEEE, 2014. 2, 3 9

    2014

  15. [23]

    Quantization and training of neural networks for efficient integer-arithmetic-only inference

    Benoit Jacob, Skirmantas Kligys, Bo Chen, Menglong Zhu, Matthew Tang, Andrew Howard, Hartwig Adam, and Dmitry Kalenichenko. Quantization and training of neural networks for efficient integer-arithmetic-only inference. In Proceed- ings of the IEEE conference on computer vision ...

    2018

  16. [24]

    Openvla: An open-source vision-language-action model

    Moo Jin Kim, Karl Pertsch, Siddharth Karamcheti, Ted Xiao, Ashwin Balakrishna, Suraj Nair, Rafael Rafailov, Ethan Fos- ter, Grace Lam, Pannag Sanketi, Quan Vuong, Thomas Kol- lar, Benjamin Burchfiel, Russ Tedrake, Dorsa Sadigh, Sergey Levine, Percy Liang, and Chelsea Finn. Ope...

    2024 arXiv

  17. [25]

    Quantizing deep convolu- tional networks for efficient inference: A whitepaper

    Raghuraman Krishnamoorthi. Quantizing deep convolu- tional networks for efficient inference: A whitepaper. arXiv preprint arXiv:1806.08342, 2018. 2

    2018 arXiv

  18. [26]

    Loftq: Lora- fine-tuning-aware quantization for large language models

    Yixiao Li, Yifan Yu, Chen Liang, Pengcheng He, Nikos Karampatziakis, Weizhu Chen, and Tuo Zhao. Loftq: Lora- fine-tuning-aware quantization for large language models. arXiv preprint arXiv:2310.08659, 2023. 3

    2023 arXiv

  19. [27]

    Cirl: Controllable imitative reinforcement learning for vision-based self-driving

    Xiaodan Liang, Tairui Wang, Luona Yang, and Eric Xing. Cirl: Controllable imitative reinforcement learning for vision-based self-driving. In Proceedings of the European conference on computer vision (ECCV) , pages 584–599,

  20. [28]

    Awq: Activation-aware weight quan- tization for llm compression and acceleration

    Ji Lin, Jiaming Tang, Haotian Tang, Shang Yang, Xingyu Dang, and Song Han. Awq: Activation-aware weight quan- tization for llm compression and acceleration. arXiv, 2023. 1, 2, 6, 12

    2023

  21. [29]

    Libero: Benchmarking knowl- edge transfer for lifelong robot learning

    Bo Liu, Yifeng Zhu, Chongkai Gao, Yihao Feng, Qiang Liu, Yuke Zhu, and Peter Stone. Libero: Benchmarking knowl- edge transfer for lifelong robot learning. Advances in Neural Information Processing Systems, 36, 2024. 2, 6, 12

    2024

  22. [30]

    Visual instruction tuning

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning. Advances in neural information processing systems, 36, 2024. 1

    2024

  23. [31]

    Nvidia tensor core pro- grammability, performance & precision

    Stefano Markidis, Steven Wei Der Chien, Erwin Laure, Ivy Bo Peng, and Jeffrey S Vetter. Nvidia tensor core pro- grammability, performance & precision. In 2018 IEEE in- ternational parallel and distributed processing symposium workshops (IPDPSW), pages 522–531. IEEE, 2018. 2

    2018

  24. [32]

    Low precision policy distillation with application to low-power, real-time sensation-cognition-action loop with neuromorphic comput- ing

    Jeffrey L Mckinstry, Davis R Barch, Deepika Bablani, Michael V Debole, Steven K Esser, Jeffrey A Kusnitz, John V Arthur, and Dharmendra S Modha. Low precision policy distillation with application to low-power, real-time sensation-cognition-action loop with neuromorphic comput-...

    2018 arXiv

  25. [33]

    Ni and The ncnn contributors

    H. Ni and The ncnn contributors. ncnn: high-performance neural network inference computing framework optimized for mobile platforms. https : / / github . com / Tencent/ncnn, 2017. 8

    2017

  26. [34]

    Nvidia tegrastats toolkit

    NVIDIA Corporation. Nvidia tegrastats toolkit. https: //docs.nvidia.com/drive/drive_os_5.1.6. 1L/nvvib_docs/index.html#page/DRIVE_OS_ Linux _ SDK _ Development _ Guide / Utilities / util_tegrastats.html. 8, 15

  27. [35]

    Open x-embodiment: Robotic learning datasets and rt-x models

    Abby O’Neill, Abdul Rehman, Abhinav Gupta, Abhiram Maddukuri, Abhishek Gupta, Abhishek Padalkar, Abraham Lee, Acorn Pooley, Agrim Gupta, Ajay Mandlekar, et al. Open x-embodiment: Robotic learning datasets and rt-x models. arXiv preprint arXiv:2310.08864, 2023. 1

    arXiv

  28. [36]

    Gpt-4 technical report

    OpenAI. Gpt-4 technical report. arXiv preprint arXiv:2303.08774, 2023. 1

    2023 arXiv

  29. [37]

    Dinov2: Learning robust visual features without supervision

    Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023. 3

    2023 arXiv

  30. [38]

    Pruning with scaled policy constraints for light- weight reinforcement learning

    Seongmin Park, Hyungmin Kim, Hyunhak Kim, and Jung- wook Choi. Pruning with scaled policy constraints for light- weight reinforcement learning. IEEE Access , 12:36055– 36065, 2024. 3

    2024

  31. [39]

    Proximal policy optimization algo- rithms

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Rad- ford, and Oleg Klimov. Proximal policy optimization algo- rithms. arXiv preprint arXiv:1707.06347, 2017. 14

    2017 arXiv

  32. [40]

    Very deep convo- lutional networks for large-scale image recognition

    Karen Simonyan and Andrew Zisserman. Very deep convo- lutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556, 2014. 2

    2014 arXiv

  33. [41]

    Algorithm-hardware co-design of adaptive floating-point encodings for resilient deep learning infer- ence

    Thierry Tambe, En-Yu Yang, Zishen Wan, Yuntian Deng, Vijay Janapa Reddi, Alexander Rush, David Brooks, and Gu-Yeon Wei. Algorithm-hardware co-design of adaptive floating-point encodings for resilient deep learning infer- ence. In 2020 57th ACM/IEEE Design Automation Confer- en...

    2020

  34. [42]

    Deepmind control suite

    Yuval Tassa, Yotam Doron, Alistair Muldal, Tom Erez, Yazhe Li, Diego de Las Casas, David Budden, Abbas Ab- dolmaleki, Josh Merel, Andrew Lefrancq, et al. Deepmind control suite. arXiv preprint arXiv:1801.00690, 2018. 12

    2018 arXiv

  35. [43]

    Carla autonomous driving leaderboard

    CARLA team. Carla autonomous driving leaderboard. https://leaderboard.carla.org/ , 2020. Ac- cessed: 2021-02-11. 12

    2020

  36. [44]

    Gemini: A family of highly capable multimodal models, 2023

    Gemini Team, Rohan Anil, et al. Gemini: A family of highly capable multimodal models, 2023. 1

    2023

  37. [45]

    MLC-LLM, 2023

    MLC team. MLC-LLM, 2023. 8

    2023

  38. [46]

    Mujoco: A physics engine for model-based control

    Emanuel Todorov, Tom Erez, and Yuval Tassa. Mujoco: A physics engine for model-based control. In 2012 IEEE/RSJ international conference on intelligent robots and systems , pages 5026–5033. IEEE, 2012. 3

    2012

  39. [47]

    Llama 2: Open foundation and fine- tuned chat models, 2023

    Hugo Touvron et al. Llama 2: Open foundation and fine- tuned chat models, 2023. 1, 3

    2023

  40. [48]

    Tinyvla: Towards fast, data-efficient vision- language-action models for robotic manipulation

    Junjie Wen, Yichen Zhu, Jinming Li, Minjie Zhu, Kun Wu, Zhiyuan Xu, Ran Cheng, Chaomin Shen, Yaxin Peng, Feifei Feng, et al. Tinyvla: Towards fast, data-efficient vision- language-action models for robotic manipulation. arXiv preprint arXiv:2409.12514, 2024. 1

    2024 arXiv

  41. [49]

    Sigmoid loss for language image pre-training

    Xiaohua Zhai, Basil Mustafa, Alexander Kolesnikov, and Lucas Beyer. Sigmoid loss for language image pre-training. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 11975–11986, 2023. 3

    2023

  42. [50]

    Fastact: A lightweight actor compression framework for fast policy learning

    Hongjie Zhang, Haoming Ma, and Zhenyu Chen. Fastact: A lightweight actor compression framework for fast policy learning. In 2023 International Joint Conference on Neural Networks (IJCNN), pages 1–8. IEEE, 2023. 3

    2023

  43. [51]

    End-to-end urban driving by imitat- ing a reinforcement learning coach

    Zhejun Zhang, Alexander Liniger, Dengxin Dai, Fisher Yu, and Luc Van Gool. End-to-end urban driving by imitat- ing a reinforcement learning coach. In Proceedings of 10 the IEEE/CVF international conference on computer vision, pages 15222–15232, 2021. 1, 2, 3, 6

    2021

  44. [52]

    Pick up the black bowl next to the ramekin and place it on the plate

    Shuchang Zhou, Yuxin Wu, Zekun Ni, Xinyu Zhou, He Wen, and Yuheng Zou. Dorefa-net: Training low bitwidth convo- lutional neural networks with low bitwidth gradients. arXiv preprint arXiv:1606.06160, 2016. 2, 3 11 A. Appendix A.1. Experiments Details A.1.1. Benchmark Details Ro...

    2016 arXiv

  45. [53]

    Training proceeds wih a learning rate of 5e-4 for a total 50,000 steps

    For wQBC, the hyper-parameter β is set at 2. Training proceeds wih a learning rate of 5e-4 for a total 50,000 steps. Autonomus Driving: In autonomous driving, tensor-wise quantization is employed for both weights and activations using LSQ [14]. The discrepancy metric D utilize...

  46. [54]

    Data Collection: The agent interacts with the environ- ment under the guidance of the quantized policy πq θold . Dur- ing this process, key information about states, actions, and rewards is recorded at each timestep and stored in an expe- rience buffer for future reinforcement...

  47. [55]

    This approach ensures that modifications remain within a permissible range, preventing any degrada- tion in the policy’s performance

    Policy Optimization: To optimize the policy, we employ Proximal Policy Optimization (PPO) [39], which utilizes a clipped objective function to carefully manage the extent of policy updates. This approach ensures that modifications remain within a permissible range, preventing ...

  48. [56]

    Integration of QBC: To enhance πq’s performance fur- ther and ensure stability in complex scenarios, we integrate the QBC as introduced in equation 3 of Sec. 4.2. QBC aids in aligning the decision-making of the quantized policy with that of a pre-trained full-precision (FP) po...

    1945

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.