How Well do Diffusion Policies Learn Kinematic Constraint Manifolds?

TL;DR

This work probes whether diffusion policies intrinsically learn kinematic constraint manifolds embedded in robot demonstration data, using a transform-locking teleoperation setup for a bimanual pick-and-place task. By systematically varying dataset size, data quality, and the curvature of the constraint manifold, it reveals that diffusion policies tend to learn a coarse, not exact, representation of the constraint manifold, with data quality playing a stronger role in constraint satisfaction and task success than data quantity. Curvature shows little to no predictive power for performance, while hardware experiments validate that the learned behavior transfers to real-world settings though robustness remains challenged by difficult sub-tasks like box pushing. The findings underscore the importance of high-quality constraint-affirming demonstrations and suggest that diffusion policies can achieve reliable task completion despite imperfect constraint learning, likely aided by low-level compliance and control.

Abstract

Diffusion policies have shown impressive results in robot imitation learning, even for tasks that require satisfaction of kinematic equality constraints. However, task performance alone is not a reliable indicator of the policy's ability to precisely learn constraints in the training data. To investigate, we analyze how well diffusion policies discover these manifolds with a case study on a bimanual pick-and-place task that encourages fulfillment of a kinematic constraint for success. We study how three factors affect trained policies: dataset size, dataset quality, and manifold curvature. Our experiments show diffusion policies learn a coarse approximation of the constraint manifold with learning affected negatively by decreases in both dataset size and quality. On the other hand, the curvature of the constraint manifold showed inconclusive correlations with both constraint satisfaction and task success. A hardware evaluation verifies the applicability of our results in the real world. Project website with additional results and visuals: https://diffusion-learns-kinematic.github.io

Figures (3)

• Figure 1: An overview of our method. We collect teleoperation data for a constrained bimanual pick-and-place task. Then, we perturb these demonstrations to generate three additional datasets that still accomplish the task, but contain increasing constraint violation. We train a policy on each of these datasets and analyze task success and constraint adherence. Lastly, we collect demonstrations for the same task on hardware, train a policy, and evaluate its performance on similar metrics.
• Figure 2: The success rates of diffusion policies trained on datasets with varying sizes and levels of perturbation. A visual depiction of each outcome is shown above. Each policy was evaluated over 200 trials; success bars include the 95% Wilson Confidence Interval for the number of full successes.
• Figure 3: An overlay of the initial conditions for the policy evaluations performed on the hardware bimanual setup.