Reinforcement Learning for Vision-Language-Action Models (RL4VLA)
RL4VLA: Reinforcement Learning for Vision-Language-Action Models
In conjunction with
Robotics: Science and Systems 2026
July 13-17, 2026
University of Technology Sydney
Sydney, Australia
Addresses how reinforcement learning can improve Vision-Language-Action models for robotics by moving beyond imitation learning limitations. Explores methods to enable policies to learn from failures, adapt to deployment distributions, and directly optimize downstream objectives rather than merely mimicking demonstrations.
More workshops at Robotics: Science and Systems 2026
- Beyond Rigidity: Deformable and Articulated Robotic Object Manipulation — Addresses robotic manipulation of non-rigid objects - deformable items like clothes and cables, plus articulated...
- Data-Centric Robotics: What Data Do Robots Really Need? — Examines the critical role of data in robot learning, asking what kinds of data matter...
- It's the Demos: A Deep Look at the Role of Demonstration Quality in Imitation-Based Robot Manipulation — Examines how demonstration quality fundamentally shapes imitation-based robot manipulation systems. Brings together researchers to establish...
- 4th Workshop on Dexterous Manipulation: Scalable Learning for Human-Level Skills — Focuses on scaling dexterous manipulation by leveraging advances in robotic hand hardware, large-scale video datasets,...
- Diffusion for Robot Learning — Explores how diffusion models can be applied to robot learning for both data generation and...
- Embodied4Arts: Robots as Creative Partners for Artistic, Expressive, and Craft Tasks — Explores how robots can learn to perform creative tasks like painting, music, and crafting by...
- Reason, Plan, Act: A Hands-On Tutorial on Embodied Agentic AI — Teaches participants how to build autonomous AI systems that combine large language models with robotics,...
- Failure Is Not the End: Towards Open-World Failure Recovery for Robots — Addresses how robots can detect, diagnose, and recover from failures when operating in open-world environments...
- Robot Planning in the Era of Foundation Models (FM4RoboPlan) — Explores how foundation models such as LLMs, vision-language models, and diffusion models are transforming robot...
- Rethinking What It Means to be Safe for Generalist Robots — Addresses how to redefine safety for generalist robots that operate across diverse tasks and environments,...
- The Geometry of Motion: Physics-Informed Structures for Learning and Control — Examines how geometric and physics-based principles such as Lie group theory and Riemannian geometry can...
- Human-centric Mobile Manipulation Workshop — Addresses how mobile manipulation robots can effectively operate alongside humans in real-world environments by developing...
- From Imitation to Certification: Learning, Reasoning, and Formal Methods for Safe Robot Manipulation — Brings together researchers from imitation learning, AI, control theory, and formal verification to address safety...
- Lab-to-Production Workshop: Toward Industrial-Grade Perception and Manipulation — Addresses the gap between robotics research and industrial deployment by examining how to translate advances...
- Workshop on Neuromorphic Robotic Systems (NeuRoSys) — Examines how neuromorphic sensing and perception - technologies inspired by biological neural systems - can...
- Open-World Navigation in the Foundation Model Era: Robustness and Failure Recovery — Examines robust navigation systems for robots in unpredictable environments by leveraging foundation models while addressing...
- Planning and Control with Imperfect Sensors and Perception — Examines how autonomous robots can make decisions when their perception systems are imperfect and uncertain,...
- Post-training for Robotics Foundation Models — Examines how robotics foundation models can improve beyond initial pretraining by collecting their own interaction...
- From Perception to Action: Representation-Centric Robot Autonomy — Examines how different map and scene representations - geometric, probabilistic, dense, implicit, or learned -...
- Mapping the Reef: Underwater 3D Reconstruction for Coral Ecosystems — Focuses on developing and advancing 3D mapping technologies for monitoring coral reef ecosystems. Brings together...
- Robot World Models — Focuses on developing world models - internal simulators that enable robots to anticipate how the...
- Sampling-Based Optimization for Robotics — Examines sampling-based techniques in robotics, exploring how methods from reinforcement learning and predictive control can...
- 3rd Workshop on Semantic Reasoning and Goal Understanding in Robotics (SemRob 2026) — Focuses on how robots achieve semantic understanding of the world to make safe decisions and...
- Closing the Sim2Real Gap: Progress and Challenges — Addresses the research challenge of bridging the gap between simulation and real-world robotics implementation. Examines...
- Bridging AI and Robotics: Towards Safer and Smarter Surgery — Explores the integration of AI and robotics in surgical applications, focusing on intelligent and autonomous...
- Tactile Sensing for Robotic Foundation Models — Explores how tactile sensing can be integrated into robotic foundation models at scale, addressing hardware...
- Embodied Intelligence Meets Foundation Models: Tactile Sensing and Soft Manipulation — Focuses on integrating tactile sensing, soft robotics, and foundation models to advance dexterous object manipulation...
- Trustworthy Embodied Foundation Models: Safety by Design versus Safety by Practice — Addresses safety challenges of deploying foundation models in robotic systems, exploring safety by design through...
- Towards Robust Execution of Long-Horizon Whole-Body Control Tasks — Addresses robot learning for complex extended tasks, focusing on deploying humanoids and manipulators on long-horizon...
- Whole-body Control and Bimanual Manipulation: Applications in Humanoids and Beyond — Brings together roboticists and machine learning researchers to discuss advances in whole-body control and bimanual...
- Beyond the Lab: Rethinking Methods for Human Behavior Monitoring and Modeling in In-the-Wild Human-Robot Interaction — Examines methodologies for observing and understanding human behavior in real-world robotic deployments, addressing the challenge...