SARL optimizes language prompt inputs to generalist vision-language-action policies through online RL to solve complex long-horizon tasks by composing existing skills.
hub Canonical reference
$\pi^{*}_{0.6}$: a VLA That Learns From Experience
Canonical reference. 76% of citing Pith papers cite this work as background.
83
Pith papers citing it
Background
76% of classified citations
abstract
We study how vision-language-action (VLA) models can improve through real-world deployments via reinforcement learning (RL). We present a general-purpose method, RL with Experience and Corrections via Advantage-conditioned Policies (RECAP), that provides for RL training of VLAs via advantage conditioning. Our method incorporates heterogeneous data into the self-improvement process, including demonstrations, data from on-policy collection, and expert teleoperated interventions provided during autonomous execution. RECAP starts by pre-training a generalist VLA with offline RL, which we call $\pi^{*}_{0.6}$, that can then be specialized to attain high performance on downstream tasks through on-robot data collection. We show that the $\pi^{*}_{0.6}$ model trained with the full RECAP method can fold laundry in real homes, reliably assemble boxes, and make espresso drinks using a professional espresso machine. On some of the hardest tasks, RECAP more than doubles task throughput and roughly halves the task failure rate.
hub tools
citation-role summary
background 30
method 4
baseline 3
citation-polarity summary
claims ledger
- abstract We study how vision-language-action (VLA) models can improve through real-world deployments via reinforcement learning (RL). We present a general-purpose method, RL with Experience and Corrections via Advantage-conditioned Policies (RECAP), that provides for RL training of VLAs via advantage conditioning. Our method incorporates heterogeneous data into the self-improvement process, including demonstrations, data from on-policy collection, and expert teleoperated interventions provided during autonomous execution. RECAP starts by pre-training a generalist VLA with offline RL, which we call $\pi
co-cited works
representative citing papers
LIBERO-Safety supplies a scalable benchmark, data-generation pipeline, and 19,664-demonstration dataset that exposes a generalization-safety tension in current VLA models where diverse training improves collision avoidance but task success stays limited by trajectory quality and semantic understandi
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.
VGenST-Bench is a new video benchmark for MLLM spatio-temporal reasoning built via generative synthesis, a multi-agent pipeline with human oversight, a 3x2x2 taxonomy, and hierarchical tasks separating perception from reasoning.
ManiSoft is a new benchmark featuring a soft-body simulator, four deformable control tasks, and an automated pipeline generating 6300 scenes with expert trajectories for training and evaluating vision-language policies on continuum robots.
RotVLA models latent actions as continuous SO(n) rotations with triplet-frame supervision and flow-matching to reach 98.2% success on LIBERO and 89.6%/88.5% on RoboTwin2.0 using a 1.7B-parameter model.
Embedding Temporal Logic (ETL) performs runtime monitoring directly in learned embedding spaces using distance-based predicates composed with temporal operators, supported by conformal calibration for reliable predicate evaluation.
A liveness-based Bellman operator enables conservative offline policy evaluation for manipulation tasks by encoding task progression and reducing truncation bias from finite horizons.
Discrete diffusion policies act as natural asynchronous executors for robotics by treating action generation as iterative unmasking, yielding higher success rates and lower computation than flow-matching real-time chunking in dynamic tasks.
PhysMem enables VLM-based robot planners to learn and verify physical properties through test-time interaction and hypothesis testing, raising success on a brick insertion task from 23% to 76%.
A2A flow matching starts action generation from prior proprioceptive actions in latent space to enable single-step high-quality predictions in robotic policies.
TouchGuide improves contact-rich robot manipulation by steering diffusion or flow-matching visuomotor policies with tactile feasibility scores from a contrastively trained Contact Physical Model.
π₀.₇ is a steerable generalist robotic model that uses rich multimodal prompts including language, subgoal images, and performance metadata to achieve out-of-the-box generalization across tasks and robot bodies.
ScoRe-Flow achieves decoupled mean-variance control in stochastic flow matching by deriving a closed-form score for drift modulation plus learned variance, yielding faster RL convergence and higher success rates on locomotion and manipulation benchmarks.
ViVa turns a video generator into a value model for robot RL that jointly forecasts future states and task value, yielding better performance on real-world box assembly when integrated with RECAP.
Action Images turn robot arm motions into interpretable multiview pixel videos, letting video backbones serve as zero-shot policies for end-to-end robot learning.
ROSA introduces shared GPU-pool serving, robotics-aware abstractions for multi-model pipelines, and factory-productivity scheduling that improves output by up to 12.06x over dedicated per-robot systems.
DART adapts VLA models to environmental shifts with one demonstration using subspace-aligned weight vector arithmetic.
Freeform Preference Learning trains language-conditioned multi-axis reward models from human pairwise preferences to produce steerable and compositional robot policies that outperform sparse and binary-preference baselines by 38 percentage points.
Z-1 uses task-wise GRPO post-training on a flow-based VLA model to reach 80.6% average success across 24 RoboCasa tasks, a 13.2-point gain over its SFT baseline.
Chronos elevates full observation history to the policy's latent state via selective SSM tokens and a Schrödinger-inspired acceleration bridge, achieving large gains on memory-dependent robot tasks with fewer parameters.
STEAM learns advantages from expert trajectories via self-supervised temporal ensemble modeling to improve policy learning on real robot tasks like bimanual folding and pick-and-place.
TAP-VLA improves VLA performance in contact-rich manipulation by visually annotating tactile shear fields onto input images, reaching 78% success versus under 50% for vision-only and other tactile methods.
EventVLA introduces foundational visual anchors and a Keyframe Evidence Memory module that predicts future keyframe probabilities from VLA embeddings to improve long-horizon task success by an average of 40% on 17 simulation and 4 real-world tasks.
citing papers explorer
-
Adapting Generalist Robot Policies with Semantic Reinforcement Learning
SARL optimizes language prompt inputs to generalist vision-language-action policies through online RL to solve complex long-horizon tasks by composing existing skills.
-
LIBERO-Safety: A Comprehensive Benchmark for Physical and Semantic Safety in Vision-Language-Action Models
LIBERO-Safety supplies a scalable benchmark, data-generation pipeline, and 19,664-demonstration dataset that exposes a generalization-safety tension in current VLA models where diverse training improves collision avoidance but task success stays limited by trajectory quality and semantic understandi
-
PhAIL: A Real-Robot VLA Benchmark and Distributional Methodology
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.
-
VGenST-Bench: A Benchmark for Spatio-Temporal Reasoning via Active Video Synthesis
VGenST-Bench is a new video benchmark for MLLM spatio-temporal reasoning built via generative synthesis, a multi-agent pipeline with human oversight, a 3x2x2 taxonomy, and hierarchical tasks separating perception from reasoning.
-
ManiSoft: Towards Vision-Language Manipulation for Soft Continuum Robotics
ManiSoft is a new benchmark featuring a soft-body simulator, four deformable control tasks, and an automated pipeline generating 6300 scenes with expert trajectories for training and evaluating vision-language policies on continuum robots.
-
RotVLA: Rotational Latent Action for Vision-Language-Action Model
RotVLA models latent actions as continuous SO(n) rotations with triplet-frame supervision and flow-matching to reach 98.2% success on LIBERO and 89.6%/88.5% on RoboTwin2.0 using a 1.7B-parameter model.
-
Runtime Monitoring of Perception-Based Autonomous Systems via Embedding Temporal Logic
Embedding Temporal Logic (ETL) performs runtime monitoring directly in learned embedding spaces using distance-based predicates composed with temporal operators, supported by conformal calibration for reliable predicate evaluation.
-
Offline Policy Evaluation for Manipulation Policies via Discounted Liveness Formulation
A liveness-based Bellman operator enables conservative offline policy evaluation for manipulation tasks by encoding task progression and reducing truncation bias from finite horizons.
-
DiscreteRTC: Discrete Diffusion Policies are Natural Asynchronous Executors
Discrete diffusion policies act as natural asynchronous executors for robotics by treating action generation as iterative unmasking, yielding higher success rates and lower computation than flow-matching real-time chunking in dynamic tasks.
-
PhysMem: Scaling Test-Time Memory for Embodied Physical Reasoning
PhysMem enables VLM-based robot planners to learn and verify physical properties through test-time interaction and hypothesis testing, raising success on a brick insertion task from 23% to 76%.
-
Action-to-Action Flow Matching
A2A flow matching starts action generation from prior proprioceptive actions in latent space to enable single-step high-quality predictions in robotic policies.
-
TouchGuide: Inference-Time Steering of Visuomotor Policies via Touch Guidance
TouchGuide improves contact-rich robot manipulation by steering diffusion or flow-matching visuomotor policies with tactile feasibility scores from a contrastively trained Contact Physical Model.
-
${\pi}_{0.7}$: a Steerable Generalist Robotic Foundation Model with Emergent Capabilities
π₀.₇ is a steerable generalist robotic model that uses rich multimodal prompts including language, subgoal images, and performance metadata to achieve out-of-the-box generalization across tasks and robot bodies.
-
ScoRe-Flow: Complete Distributional Control via Score-Based Reinforcement Learning for Flow Matching
ScoRe-Flow achieves decoupled mean-variance control in stochastic flow matching by deriving a closed-form score for drift modulation plus learned variance, yielding faster RL convergence and higher success rates on locomotion and manipulation benchmarks.
-
ViVa: A Video-Generative Value Model for Robot Reinforcement Learning
ViVa turns a video generator into a value model for robot RL that jointly forecasts future states and task value, yielding better performance on real-world box assembly when integrated with RECAP.
-
Action Images: End-to-End Policy Learning via Multiview Video Generation
Action Images turn robot arm motions into interpretable multiview pixel videos, letting video backbones serve as zero-shot policies for end-to-end robot learning.
-
ROSA: A Robotics Foundation Model Serving System for Robot Factories
ROSA introduces shared GPU-pool serving, robotics-aware abstractions for multi-model pipelines, and factory-productivity scheduling that improves output by up to 12.06x over dedicated per-robot systems.
-
Domain Arithmetic: One-Shot VLA Adaptation under Environmental Shifts
DART adapts VLA models to environmental shifts with one demonstration using subspace-aligned weight vector arithmetic.
-
Freeform Preference Learning for Robotic Manipulation
Freeform Preference Learning trains language-conditioned multi-axis reward models from human pairwise preferences to produce steerable and compositional robot policies that outperform sparse and binary-preference baselines by 38 percentage points.
-
Z-1: Efficient Reinforcement Learning for Vision-Language-Action Models
Z-1 uses task-wise GRPO post-training on a flow-based VLA model to reach 80.6% average success across 24 RoboCasa tasks, a 13.2-point gain over its SFT baseline.
-
Chronos: A Physics-Informed Full-History Framework for Non-Markovian Long-Horizon Manipulation
Chronos elevates full observation history to the policy's latent state via selective SSM tokens and a Schrödinger-inspired acceleration bridge, achieving large gains on memory-dependent robot tasks with fewer parameters.
-
STEAM: Self-Supervised Temporal Ensemble Advantage Modeling for Real-World Robot Learning
STEAM learns advantages from expert trajectories via self-supervised temporal ensemble modeling to improve policy learning on real robot tasks like bimanual folding and pick-and-place.
-
TAP-VLA: Tactile Annotation Prompting for Vision Language Action Models
TAP-VLA improves VLA performance in contact-rich manipulation by visually annotating tactile shear fields onto input images, reaching 78% success versus under 50% for vision-only and other tactile methods.
-
EventVLA: Event-Driven Visual Evidence Memory for Long-Horizon Vision-Language-Action Policies
EventVLA introduces foundational visual anchors and a Keyframe Evidence Memory module that predicts future keyframe probabilities from VLA embeddings to improve long-horizon task success by an average of 40% on 17 simulation and 4 real-world tasks.
-
What Matters When Cotraining Robot Manipulation Policies on Everyday Human Videos?
Cotraining on 532 everyday human videos with accurate hand labels improves robot policies by 29.7% when networks specialize to human versus robot embodiments.
-
FlowPRO: Reward-Free Reinforced Fine-Tuning of Flow-Matching VLAs via Proximalized Preference Optimization
FlowPRO applies proximalized preference optimization to flow-matching VLAs with intervention-rollback data to reach higher success rates on long-horizon bimanual tasks without rewards or critics.
-
Set-Supervised Diffusion Policy: Learning Action-Chunking Diffusion through Corrections
SDP constructs sets of desired action-chunks from human correction pairs and trains diffusion policies to align with those sets, yielding better performance and robustness than standard behavior cloning on robotic tasks.
-
Hide-and-Seek in Trajectories: Discovering Failure Signals for VLA Runtime Monitoring
Hide-and-Seek uses contrastive objectives on trajectories to localize failure signals in VLA models from trajectory-level supervision alone.
-
Feat2Go: Visual Feature-Grounded Value Estimation for Embodied Reinforcement Learning
Feat2Go uses patch-level similarity from a visual world model and trend-based clustering to create progress targets for training value models that improve reward shaping in embodied RL for VLA policies, yielding large gains on ManiSkill3 and RoboTwin benchmarks.
-
ParkingWorld: End-to-End Autonomous Parking Reinforcement Learning from Corrective Experience in 3DGS Simulation
CIL-SERL integrates a hierarchical replay buffer with human corrective interventions into RL for autonomous parking in 3DGS simulation, reporting gains in success rate, efficiency, and safety on sim and real vehicle.
-
Beyond Action Residuals: Real-World Robot Policy Steering via Bottleneck Latent Reinforcement Learning
ZPRL adapts frozen flow-matching imitation policies via RL perturbations on a task-relevant bottleneck latent, yielding 33.7% higher average success on four real-world manipulation tasks than action-residual baselines.
-
Hand-in-the-Loop: Improving VLA Policies for Dexterous Manipulation via Seamless Hand-Arm Intervention
HandITL enables seamless human intervention in VLA policies for bimanual dexterous manipulation, cutting jitter by 99.8% and improving refined policies by 19% over standard teleoperation.
-
Reinforcing VLAs in Task-Agnostic World Models
RAW-Dream disentangles world-model learning from task data by using a pre-trained task-agnostic world model and VLM rewards, with dual-noise filtering, to enable zero-shot VLA adaptation in simulation and real settings.
-
TMRL: Diffusion Timestep-Modulated Pretraining Enables Exploration for Efficient Policy Finetuning
TMRL bridges behavioral cloning pretraining and RL finetuning via diffusion noise and timestep modulation to enable controlled exploration, improving sample efficiency and enabling real-world robot training in under one hour.
-
Learning While Deploying: Fleet-Scale Reinforcement Learning for Generalist Robot Policies
LWD is a fleet-scale offline-to-online RL framework that continually improves pretrained VLA policies using autonomous rollouts and human interventions, reaching 95% average success on real-world manipulation tasks.
-
LaST-R1: Reinforcing Robotic Manipulation via Adaptive Physical Latent Reasoning
LaST-R1 introduces a RL post-training method called LAPO that optimizes latent Chain-of-Thought reasoning in vision-language-action models, yielding 99.9% success on LIBERO and up to 22.5% real-world gains.
-
Multi-View Video Diffusion Policy: A 3D Spatio-Temporal-Aware Video Action Model
MV-VDP jointly predicts multi-view RGB and heatmap videos via diffusion to achieve data-efficient, robust robotic manipulation policies.
-
ARM: Advantage Reward Modeling for Long-Horizon Manipulation
ARM trains reward models on Progressive/Regressive/Stagnant labels to enable adaptive reweighting in offline RL, reaching 99.4% success on towel-folding with minimal human intervention.
-
Open-Loop Planning, Closed-Loop Verification: Speculative Verification for VLA
SV-VLA uses infrequent heavy VLA planning of action chunks plus a lightweight closed-loop verifier to achieve both efficiency and robustness in dynamic robot control.
-
FASTER: Rethinking Real-Time Flow VLAs
FASTER adds a Horizon-Aware Schedule to flow VLAs that compresses immediate-action denoising to one step while keeping long-horizon trajectory quality, lowering real-robot reaction latency.
-
Simulation Distillation: Pretraining World Models in Simulation for Rapid Real-World Adaptation
SimDist pretrains world models in simulation and adapts them to real-world robots by updating only the latent dynamics model, enabling rapid improvement on contact-rich tasks where prior methods fail.
-
CoFL: Continuous Flow Fields for Language-Conditioned Navigation
CoFL learns continuous flow fields from BEV images and language instructions to generate navigation trajectories, outperforming modular VLM planners and trajectory policies on unseen scenes.
-
Robometer: Scaling General-Purpose Robotic Reward Models via Trajectory Comparisons
Robometer combines intra-trajectory progress supervision with inter-trajectory preference supervision on a 1M-trajectory dataset to learn more generalizable robotic reward functions than prior methods.
-
VLANeXt: Recipes for Building Strong VLA Models
VLANeXt distills 12 design insights from a unified VLA study into a model that outperforms prior methods on LIBERO benchmarks while releasing code for further exploration.
-
RISE: Self-Improving Robot Policy with Compositional World Model
RISE combines a controllable dynamics model and progress value model into a closed-loop self-improving pipeline that updates robot policies entirely in imagination, reporting over 35% absolute gains on three real-world tasks.
-
Self-Supervised Bootstrapping of Action-Predictive Embodied Reasoning
R&B-EnCoRe uses self-supervised importance-weighted variational inference to distill action-predictive reasoning datasets that improve VLA performance on manipulation, navigation, and driving tasks without external verifiers.
-
PALM: Progress-Aware Policy Learning via Affordance Reasoning for Long-Horizon Robotic Manipulation
PALM improves long-horizon robotic manipulation success by distilling affordance representations for object interaction and predicting within-subtask progress in a VLA model.
-
A Survey on Vision-Language-Action Models for Embodied AI
This is the first survey on vision-language-action models, providing a taxonomy across three lines, plus summaries of datasets, simulators, benchmarks, challenges, and future directions in embodied AI.
-
Unified Noise Steering for Efficient Human-Guided VLA Adaptation
UniSteer unifies human corrective actions and noise-space RL for VLA adaptation by inverting actions to noise targets, raising success rates from 20% to 90% in 66 minutes across four real-world manipulation tasks.
-
RePO-VLA: Recovery-Driven Policy Optimization for Vision-Language-Action Models
RePO-VLA raises average adversarial success rates in VLA manipulation from 20% to 75% by using recovery-aware initialization, a progress-aware semantic value function, and value-conditioned refinement on success and corrective trajectories.