Vision language helps robots avoid task specific object hazards in 3d space
Grounding Object-to-Object Hazard Reasoning into 3D Risk Fields for Task-Conditioned Safety
Robotics
Summary
Robots need to be safe not just by avoiding bumping into things, but by understanding what tasks they are doing and which objects might cause hazards because of how they interact. The authors created a system that uses vision and language models to look at a scene and know if a task is too risky to do. If it is safe enough, the system spots pairs of objects that might cause problems and marks the risky spots around them in 3D. The robot then adjusts its movements to steer clear of these danger zones, making manipulation tasks safer. They tested this idea both in computer simulations and with real robots.
What this means in practice
- •For robotics engineers: Build manipulation robots that dynamically avoid hazards based on task instructions and visual scene understanding in 3D.
- •For industrial automation integrators: Enhance factory robots to reject unsafe tasks and prevent object-to-object hazards during complex assembly or handling.
Authors
Mariana Sosa Guzmán, Yide Song, Amin Abyaneh, Gregory Dudek, Hsiu-Chin Lin
Abstract
Safety in robotics does not solely depend on geometric constraints, but also on semantics: the given task, environment properties, and interactions between objects in the scene. We present a framework based on vision-language models (VLMs) to infer task- and scene-dependent safety constraints for robot manipulation. We provide the VLM a structured safety tool based on an RGB-D observation and a natural-language task instruction that first decides whether the task should be rejected since its execution may result in a non-recoverable hazardous outcome. Otherwise, it identifies preventable object-pair hazards and specifies the spatial relations that must be avoided. Open-vocabulary segmentation and depth information are then used to ground the detected objects in 3D and construct continuous risk fields that represent the corresponding hazardous regions. These fields are integrated with an obstacle avoidance module that corrects nominal robot motions during execution while accounting for the risk field. We evaluate the framework on preventable hazards, tasks requiring rejection, and safe variants in simulation and the real-world. Project website: https://vlm-hazard-riskfield.github.io