Household robots struggle to adjust motion by obstacle type safely
Benchmarking Obstacle-Conditioned Safety in Household Robot Navigation
RoboticsArtificial IntelligenceComputer Vision and Pattern Recognition
Summary
Robots that move safely around household obstacles need to adjust how closely and quickly they pass different objects. The authors created a new test, OSMA-Bench, to see if robots change their movements depending on the kind of obstacle near them. They found current robot motion policies can tell what kind of obstacle is ahead but do not change their speed or path as carefully as needed. Adding a step where the robot figures out safety parameters improved this behavior for some models. This work highlights the challenge of turning what a robot understands about obstacles into safe, smooth motions.
What this means in practice
- •For robotics engineers: Improve robot navigation systems by integrating obstacle safety distinctions into motion adjustment modules for safer household operation.
- •For autonomous vehicle developers: Incorporate semantic safety sensitivity measures to better adapt vehicle speed and clearance dynamically to different obstacle types.
Authors
Hyunwoong Bae, Jisu Han, Jiyeon Choi, Dohoon Kim, Taesup Moon
Abstract
Safe household navigation requires robots to regulate clearance and speed according to what an obstacle is and how close it is, without unnecessary conservatism. Yet existing evaluations do not test whether semantic safety distinctions are reflected in continuous motion. To that end, we introduce OSMA-Bench, an Obstacle-Conditioned Safe Motion Adaptation Benchmark built on spatially matched scenario sets that differ only in obstacle type. We also propose the \emph{Semantic Sensitivity Index} (SSI), which measures whether clearance and distance-dependent speed and smoothness follow the caution ordering of obstacles. Evaluating VLM-based and vision-language-action (VLA) policies on the benchmark, we find that neither consistently adapts its motion to obstacle safety, even with explicit safety prompts. Our analysis shows that the policies can correctly distinguish obstacles by safety relevance but fail to reflect this distinction in the motion parameters or actions they produce. For VLM-based policies, which recognize the obstacle and then generate a navigation program, adding an explicit parameter-inference step between these two stages improves adaptation. These findings expose a gap in \emph{semantic-to-motion grounding}, highlighting the need to translate semantic caution into appropriately safe motion.