Moving On, Even When You�re Broken: Fail-Active Trajectory Generation Via Diffusion Policies Conditioned on Embodiment and Task
Gilberto Briscoe-Martinez, Yaashia Gautam, Rahul Shetty, Anuj Pasricha, Marco Nicotra, Alessandro Roncone
AI summary
Problem
Current safety standards force robots to halt upon detecting faults, causing prolonged downtime, while existing fail-active methods cannot generalize across arbitrary multi-joint failures or multiple manipulation primitives.
Approach
DEFT conditions a diffusion model on structured embeddings of joint-level failures and task constraints to synthesize feasible, adaptive trajectories online without retraining or policy switching.
Key results
- 99.5% success rate for unconstrained motions versus RRT’s 42.4%
- 46.4% success rate for constrained motions versus differential IK’s 30.9%
- Robust zero-shot generalization to unseen failure configurations
- Successful real-world execution of multi-step tasks under induced joint failures
Why it matters
Provides a scalable, unified framework for maintaining robotic autonomy and task completion during hardware degradation, critical for long-duration space, industrial, and hazardous environment operations.
Abstract
Robot failure is detrimental and disruptive, often requiring human intervention to recover. Our vision is fail- active operation, allowing robots to safely complete their tasks even when damaged. Focusing on actuation failures, we introduce DEFT, a diffusion-based trajectory generator condi- tioned on the robot’s current embodiment and task constraints. DEFT generalizes across failure types, supports constrained and unconstrained motions, and enables task completion under ar- bitrary failure. We evaluate DEFT in both simulation and real- world scenarios using a 7-DoF robotic arm. DEFT outperforms its baselines over thousands of failure conditions, achieving a 99.5% success rate for unconstrained motions versus RRT’s 42.4%, and 46.4% for constrained motions versus differential IK’s 30.9%. Furthermore, DEFT demonstrates robust zero- shot generalization by maintaining performance on failure conditions unseen during training. Finally, we perform real- world evaluations on two multi-step tasks, drawer manipula- tion and whiteboard erasing. These experiments demonstrate DEFT succeeding on tasks where classical methods fail. Our results show that DEFT achieves fail-active manipulation across arbitrary failure configurations and real-world deployments.