VLA models from VLM adaptation can be pruned 12-30% via multi-module joint scheme based on divergence signals while keeping ~90% performance on LIBERO without post-pruning recovery, unlike standard criteria that collapse.
super hub Canonical reference
SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics
Canonical reference. 71% of citing Pith papers cite this work as background.
146
Pith papers citing it
Background
71% of classified citations
abstract
Vision-language models (VLMs) pretrained on large-scale multimodal datasets encode rich visual and linguistic knowledge, making them a strong foundation for robotics. Rather than training robotic policies from scratch, recent approaches adapt VLMs into vision-language-action (VLA) models that enable natural language-driven perception and control. However, existing VLAs are typically massive--often with billions of parameters--leading to high training costs and limited real-world deployability. Moreover, they rely on academic and industrial datasets, overlooking the growing availability of community-collected data from affordable robotic platforms. In this work, we present SmolVLA, a small, efficient, and community-driven VLA that drastically reduces both training and inference costs, while retaining competitive performance. SmolVLA is designed to be trained on a single GPU and deployed on consumer-grade GPUs or even CPUs. To further improve responsiveness, we introduce an asynchronous inference stack decoupling perception and action prediction from action execution, allowing higher control rates with chunked action generation. Despite its compact size, SmolVLA achieves performance comparable to VLAs that are 10x larger. We evaluate SmolVLA on a range of both simulated as well as real-world robotic benchmarks and release all code, pretrained models, and training data.
hub tools
citation-role summary
background 25
baseline 8
method 1
citation-polarity summary
claims ledger
- abstract Vision-language models (VLMs) pretrained on large-scale multimodal datasets encode rich visual and linguistic knowledge, making them a strong foundation for robotics. Rather than training robotic policies from scratch, recent approaches adapt VLMs into vision-language-action (VLA) models that enable natural language-driven perception and control. However, existing VLAs are typically massive--often with billions of parameters--leading to high training costs and limited real-world deployability. Moreover, they rely on academic and industrial datasets, overlooking the growing availability of comm
authors
co-cited works
representative citing papers
SurgVLA-Bench supplies a hierarchical task taxonomy and multi-dimensional evaluation framework for VLA models in laparoscopic robotics simulation, showing autoregressive models excel at semantics while flow-matching models achieve higher precision but all fall short due to endoscopic view constraint
FAFM performs flow matching in the frequency domain using DCT on action sequences to produce continuous temporally consistent robotic actions with a Sobolev-style smoothness regularizer.
EquiVLA is the first general framework for end-to-end SO(2)-equivariant VLA models using EquiPerceptor and EquiActor modules, reporting improved success rates on LIBERO, CALVIN, and real-robot benchmarks.
PAINT reframes asynchronous flow-based action chunking as an initial noise selection problem solved via backward Euler inversion and a repainting rule.
Act2Answer protocol reveals VLA models retain simple concepts but show larger gaps on complex semantics than source VLMs, with VQA co-training linked to better retention and knowledge signals peaking in middle layers.
MuseVLA adds on-demand sensor selection via tokens and converts readings into grounded sensor images for multimodal fusion, reporting 80.6% average success on real-robot dexterous tasks that need non-visual sensing.
LeaP introduces a learnable proprioception-conditioned diagonal Gaussian source prior for generative robot policies, raising average success rates on 15 RoboTwin tasks from baselines by 6.5-25.5 points.
A prompt-only attack called command-preserving trajectory redirection can steer VLA robot behavior to attacker-chosen physical outcomes while the text still appears to match the intended task.
World models introduce a stealthy poisoning vector into robot learning pipelines where malicious prompts or dynamics in teleoperated data activate only during synthetic trajectory generation, enabling backdoors in downstream policies.
RLDT fine-tunes pretrained flow-matching policies for continuous control by aligning them to a max-entropy RL transport field constructed via SVGD, using expected-target estimation for stable multi-step updates.
PhAIL provides an open benchmark and distributional evaluation method for real-robot VLA policies using time-to-success CDF, HRT scoring, and KS significance tests.
Benchmarking ACT, Diffusion Policy, SmolVLA, and π0 on suture following yields 50-75% success under ideal conditions and 92% stitch completion with π0 in a surgeon-robot trial.
VLA architectures exhibit architecture-specific failure signatures at the motor-command level, with direction reversal as a universal predictor and velocity monitoring ineffective for continuous models.
JOPAT jointly models pixels, point tracks, and actions in a diffusion transformer and reports gains over pixel-only baselines on long-horizon robot tasks with occlusion and off-screen motion.
The paper identifies distinct failure mechanisms: excessive posterior-prior regularization erases mode information in latent policies, while smooth base-to-action maps limit mode coverage in generative policies.
Pion modifies Muon's Newton-Schulz iterations into a controllable high-pass filter that anchors dominant singular values at 1 while suppressing noisy tails, outperforming Muon and AdamW in VLA and RLVR regimes.
A new speculative inference system speeds up diffusion VLAs to 19.1 ms average latency (3.04x faster) on LIBERO by replacing most full 58 ms inferences with 7.8 ms draft rounds while preserving task performance.
Test-time sparsity with a parallel pipeline and omnidirectional feature reuse accelerates action diffusion by 5x to 47.5 Hz while cutting FLOPs 92% with no performance loss.
MoLA infers a mixture of latent actions from generated future videos via modality-aware inverse dynamics models to improve robot manipulation policies.
Premover enables VLA policies to act on partial instructions by precomputing focus maps from intermediate backbone layers, reducing wall-clock time 13.6 percent on LIBERO while preserving 95 percent success rate.
GridS is a plug-and-play differentiable module for geometry-aware visual token resampling in VLA models that achieves under 10% token retention and 76% FLOPs reduction with no success-rate loss.
Pace-and-Path Correction decomposes a quadratic cost minimization into orthogonal pace and path channels to correct chunked actions in VLA models, raising success rates by up to 28.8% in dynamic settings.
VLATIM benchmark reveals large VLMs excel at high-level planning in physics puzzles but struggle with precise visual grounding and mouse control, so they lack human-like problem-solving capabilities.
citing papers explorer
-
SurgVLA-Bench: Towards Evaluating Vision-Language-Action Models for Laparoscopic Surgical Robotics
SurgVLA-Bench supplies a hierarchical task taxonomy and multi-dimensional evaluation framework for VLA models in laparoscopic robotics simulation, showing autoregressive models excel at semantics while flow-matching models achieve higher precision but all fall short due to endoscopic view constraint
-
Do Vision-Language-Models show human-like logical problem-solving capability in point and click puzzle games?
VLATIM benchmark reveals large VLMs excel at high-level planning in physics puzzles but struggle with precise visual grounding and mouse control, so they lack human-like problem-solving capabilities.
-
LoopVLA: Learning Sufficiency in Recurrent Refinement for Vision-Language-Action Models
LoopVLA adds recurrent refinement and learned sufficiency estimation to VLA models, cutting parameters 45% and raising throughput 1.7x while matching baseline task success on LIBERO and VLA-Arena.
-
RL-VLA$^3$: A Flexible and Asynchronous Reinforcement Learning Framework for VLA Training
RL-VLA³ is an asynchronous RL framework for VLA training that delivers up to 85.2% higher throughput than synchronous baselines while preserving identical sample efficiency and scaling to 256 GPUs.
-
Memory as a Wasting Asset: Pricing Flash Endurance for Embodied Agents, and the Limits of Doing So
Flash endurance is priced via shadow price η making placement cost-optimal for any sign of value-write correlation χ, with χ positive only in recurrent long-horizon manipulation and the budget binding only on low-endurance commodity hardware.
-
AEGIS: A Backup Reflex for Physical AI
AEGIS uses activation probes for early-warning detection of high-risk steps in weak policies and selectively escalates to stronger policies, recovering 10.1% of lost trajectories on LIBERO-Spatial while activating the strong policy on only 38% of steps.
-
Learning Bilevel Policies over Symbolic World Models for Long-Horizon Planning
BISON learns bilevel policies over symbolic world models to generalize long-horizon robotic planning beyond VLA and end-to-end baselines while remaining efficient even at 10,000-object scale.
-
VGAS: Value-Guided Action-Chunk Selection for Few-Shot Vision-Language-Action Adaptation
VGAS uses best-of-N selection with a geometrically grounded critic and explicit regularization to improve success rates of few-shot VLA policies under limited data and distribution shifts.
-
Aligning Perception, Reasoning, Modeling and Interaction: A Survey on Physical AI
A survey of physical AI that distinguishes theoretical physics reasoning from applied understanding and synthesizes advances in symbolic reasoning, embodied systems, and generative models to advocate for physics-grounded world models.