CSF: Contextual Safety Filtering for Motion Generators
Contextual safety filtering grounds language rules in trajectories and cuts motion danger events by up to 90%.
Researchers introduce Contextual Safety Filtering, a training-free method that grounds natural-language safety rules in safe and unsafe reference trajectories from a text-conditioned motion generator. An affine safety value from those trajectories is enforced by a control-barrier-function quadratic program. Across four pretrained generators, the filter activated the intended rules in all explicit and scene-triggered unsafe cases, reduced the danger-event rate by up to 90%, and preserved 88-100% of benign motions. A complete system on a Unitree G1 blocked unsafe motions involving people and objects.
- Training-free filter uses safe and unsafe generator trajectories as references.
- A tracking CBF-QP enforces an affine safety value for each active rule.
- Evaluated on four pretrained generators with different architectures.
- Danger-event rate fell by up to 90% while 88-100% of benign motions remained.
- Unitree G1 trials blocked unsafe interactions with humans and objects.
Full article159 words · extracted from arxiv.org · click to collapse
Text-conditioned motion generators produce trackable whole-body motion, but they have no notion of scene-dependent safety: the same action may target an object or a person. Existing safeguards either inspect the prompt, require labeled motion data, or enforce geometric constraints; therefore, they do not directly account for how scene context changes a motion's meaning. We introduce contextual safety filtering (CSF), a training-free filter that grounds natural-language safety rules in safe and unsafe reference trajectories produced by the generator. For each active rule, safe and unsafe reference trajectories define an affine safety value that a safe reference tracking CBF-QP enforces. Across four pretrained generators with different architectures, CSF activates the intended rules in all explicit and scene-triggered unsafe cases and reduces the danger-event rate by up to 90%, while preserving 88-100% of benign motions. We demonstrate the complete system on a real-world Unitree G1, where it successfully prevents unsafe motions in a variety of scenarios, including interactions with humans and objects.
Text extracted automatically; images, tables and formatting may be missing. Original: https://arxiv.org/abs/2610.12467