Robot policies learn safe actions from simple feasibility checks
Learning Safe Flow Matching Robot Policies from Binary Feasibility Oracles
RoboticsArtificial Intelligence
Summary
Robots must follow safety rules like avoiding obstacles, but sometimes it's hard to define these rules precisely. The authors created a method that only needs simple yes-no safety tests to teach robots how to move safely. Their approach adjusts the robot’s plans in real time to avoid unsafe moves and learns from these adjustments. This lets robots safely complete tasks without needing detailed safety models.
What this means in practice
- •For industrial robot programmers: Create robot control policies that ensure safety without needing explicit mathematical safety models, using only simple binary checks.
- •For autonomous warehouse operators: Deploy robots that reliably avoid collisions and joint violations during manipulation tasks by integrating feasible safety checks in learning.
Authors
Jianming Ma, Rongjun Jin, Yizhi Chen, Yue Gao
Abstract
Robot foundation models increasingly rely on flow-based action generation for general-purpose manipulation, yet the generated actions are not guaranteed to satisfy physical constraints such as joint limits and collision avoidance. This problem is especially challenging when the feasible set cannot be described analytically and is accessible only through black-box feasibility oracles. We study this practical oracle-only setting and propose an oracle-guided differentiable safeguard for safe flow-matching robot policies. The safeguard uses binary feasibility queries to correct unsafe updates in the forward pass, while extracting local boundary information to provide constraint-aware learning signals in the backward pass. Combined with discrete-time flow matching, the same safeguard is used during training and inference, enabling forward--backward aligned policy learning without explicit or differentiable constraint models. Experiments on obstacle-aware robotic manipulation demonstrate 100% constraint satisfaction while preserving task performance.