Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Moving On, Even When You're Broken: Fail-Active Trajectory Generation via Diffusion Policies Conditioned on Embodiment and Task

Gilberto G. Briscoe-Martinez, Yaashia Gautam, Rahul Shetty, Anuj Pasricha, Marco M. Nicotra, Alessandro Roncone

TL;DR

This work addresses the challenge of maintaining manipulation capabilities under actuation failures by introducing DEFT, a diffusion-based trajectory generator conditioned on embodiment and task constraints. DEFT uses an embodiment vector $oldsymbol{\xi}$ and a task constraint vector $oldsymbol{\tau}$, along with start–goal inpainting and endpoint clamping, to synthesize feasible joint-space trajectories without policy switching or retraining. Its key contributions include per-joint failure encoding, FiLM-based conditioning for both embodiment and task constraints, and a data-generation pipeline that yields both constrained and unconstrained motion primitives. Across extensive simulation and real-world experiments on a 7-DoF Panda arm, DEFT outperformed classical planners and ablations, generalizing to unseen failures and enabling perfect task completion on long-horizon tasks such as drawer manipulation and whiteboard erasing. The results demonstrate the practical impact of fail-active diffusion policies for robust autonomous operation under hardware faults, with broad potential for cross-embodiment transfer and real-time fault adaptation in future work.

Abstract

Robot failure is detrimental and disruptive, often requiring human intervention to recover. Maintaining safe operation under impairment to achieve task completion, i.e. fail-active operation, is our target. Focusing on actuation failures, we introduce DEFT, a diffusion-based trajectory generator conditioned on the robot's current embodiment and task constraints. DEFT generalizes across failure types, supports constrained and unconstrained motions, and enables task completion under arbitrary failure. We evaluated DEFT in both simulation and real-world scenarios using a 7-DoF robotic arm. In simulation over thousands of joint-failure cases across multiple tasks, DEFT outperformed the baseline by up to 2 times. On failures unseen during training, it continued to outperform the baseline, indicating robust generalization in simulation. Further, we performed real-world evaluations on two multi-step tasks, drawer manipulation and whiteboard erasing. These experiments demonstrated DEFT succeeding on tasks where classical methods failed. Our results show that DEFT achieves fail-active manipulation across arbitrary failure configurations and real-world deployments.

Moving On, Even When You're Broken: Fail-Active Trajectory Generation via Diffusion Policies Conditioned on Embodiment and Task

TL;DR

This work addresses the challenge of maintaining manipulation capabilities under actuation failures by introducing DEFT, a diffusion-based trajectory generator conditioned on embodiment and task constraints. DEFT uses an embodiment vector and a task constraint vector , along with start–goal inpainting and endpoint clamping, to synthesize feasible joint-space trajectories without policy switching or retraining. Its key contributions include per-joint failure encoding, FiLM-based conditioning for both embodiment and task constraints, and a data-generation pipeline that yields both constrained and unconstrained motion primitives. Across extensive simulation and real-world experiments on a 7-DoF Panda arm, DEFT outperformed classical planners and ablations, generalizing to unseen failures and enabling perfect task completion on long-horizon tasks such as drawer manipulation and whiteboard erasing. The results demonstrate the practical impact of fail-active diffusion policies for robust autonomous operation under hardware faults, with broad potential for cross-embodiment transfer and real-time fault adaptation in future work.

Abstract

Robot failure is detrimental and disruptive, often requiring human intervention to recover. Maintaining safe operation under impairment to achieve task completion, i.e. fail-active operation, is our target. Focusing on actuation failures, we introduce DEFT, a diffusion-based trajectory generator conditioned on the robot's current embodiment and task constraints. DEFT generalizes across failure types, supports constrained and unconstrained motions, and enables task completion under arbitrary failure. We evaluated DEFT in both simulation and real-world scenarios using a 7-DoF robotic arm. In simulation over thousands of joint-failure cases across multiple tasks, DEFT outperformed the baseline by up to 2 times. On failures unseen during training, it continued to outperform the baseline, indicating robust generalization in simulation. Further, we performed real-world evaluations on two multi-step tasks, drawer manipulation and whiteboard erasing. These experiments demonstrated DEFT succeeding on tasks where classical methods failed. Our results show that DEFT achieves fail-active manipulation across arbitrary failure configurations and real-world deployments.
Paper Structure (33 sections, 5 equations, 4 figures, 3 tables, 2 algorithms)

This paper contains 33 sections, 5 equations, 4 figures, 3 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Failure-Aware Robot Control
  4. Diffusion Models for Trajectory Generation
  5. Methods
  6. DEFT
  7. Conditioning Approaches
  8. Embodiment Conditioning
  9. Constraint Conditioning
  10. Data Generation Approaches
  11. Start and Goal Sampling
  12. Trajectory Construction
  13. Failure Condition Sampling
  14. Task Validity Filtering
  15. Experimental Evaluation
  16. ...and 18 more sections

Figures (4)

  • Figure 1: DEFT takes as input a structured embodiment encoding representing joint-level failures and a constraint encoding then synthesizes a feasible robot trajectory via a diffusion model. After experiencing a failure, a given task will likely need to be completed in a different manner. In this figure, the pick-and-place segment is no longer feasible, given the failure condition, and the robot must now push the object to a graspable state to complete the task.
  • Figure 2: Overview of DEFT. a) joint-level embodiment constraints are encoded into a structured representation capturing failure constrained joint position and velocity limits. This embodiment encoding is processed by a multilayer perceptron (MLP) and used to condition the diffusion model. b) Task-specific constraints (e.g. unconstrained or constrained) are represented as one-hot vectors and concatenated with the embodiment encoding to constrained the generative process. Given these conditioning signals and start-goal joint configurations, the model generates feasible joint-space trajectories adapted explicitly to both the robot's degraded embodiment and task requirements. At each step of the denoising process the predicted joint values are clamped to the failure-induced robot joint limits.
  • Figure 3: Graphic description of the constraint definitions used in \ref{['sec:sim-analysis']}. The unconstrained primitive corresponds to a feasible trajectory between two points and includes unconstrained manipulation where the object is rigidly secured to the end-effector. The constrained corresponds to planar, approximately straight-line motion with minimal end-effector orientation change.
  • Figure 4: Left (Erasing): The robot grasps an eraser from the table and performs back-and-forth sweeps on the whiteboard to remove text. Right (Drawer): The robot opens the drawer, pushes an object to a graspable pose, grasps it, places it inside, and then closes the drawer. Observed failure modes included, for erasing, incomplete wipes or drops when constraint-consistent motion was not obeyed (excess or insufficient normal force) and missed picks when, without inpainting, the policy failed to reach the prescribed start and goal poses; and for the drawer, failure to open when the constraint motion was ignored, inability to find a feasible plan under task constraints, and drops when, without inpainting, the target positions were not respected.