EDMP: Ensemble-of-costs-guided Diffusion for Motion Planning

TL;DR

EDMP, an Ensemble-of-costs-guided Diffusion for Motion Planning that aims to combine the strengths of classical and deep-learning-based motion planning, is proposed, which performs comparably with SOTA deep-learning-based methods while retaining the generalization capabilities primarily associated with classical planners.

Abstract

Classical motion planning for robotic manipulation includes a set of general algorithms that aim to minimize a scene-specific cost of executing a given plan. This approach offers remarkable adaptability, as they can be directly used off-the-shelf for any new scene without needing specific training datasets. However, without a prior understanding of what diverse valid trajectories are and without specially designed cost functions for a given scene, the overall solutions tend to have low success rates. While deep-learning-based algorithms tremendously improve success rates, they are much harder to adopt without specialized training datasets. We propose EDMP, an Ensemble-of-costs-guided Diffusion for Motion Planning that aims to combine the strengths of classical and deep-learning-based motion planning. Our diffusion-based network is trained on a set of diverse kinematically valid trajectories. Like classical planning, for any new scene at the time of inference, we compute scene-specific costs such as "collision cost" and guide the diffusion to generate valid trajectories that satisfy the scene-specific constraints. Further, instead of a single cost function that may be insufficient in capturing diversity across scenes, we use an ensemble of costs to guide the diffusion process, significantly improving the success rate compared to classical planners. EDMP performs comparably with SOTA deep-learning-based methods while retaining the generalization capabilities primarily associated with classical planners.

Figures (8)

• Figure 1: Importance of Ensemble. Using a single cost function alongside a diffusion model may risk collision or generate sub-optimal trajectories. EDMP utilizes an ensemble of cost functions with varying hyperparameters along with the diffusion model. This ensures that an optimal solution can be found even in complex scenes.
• Figure 2: Architecture. EDMP leverages a diffusion model alongside a 12-member cost ensemble. The diffusion model denoises a batch of trajectories while each cost in the ensemble guides a specific slice of the batch. The cost function calculates the intersection or swept volume from robot and environment bounding boxes. The trajectory with minimum swept volume is chosen as the final solution
• Figure 3: (a) Gradient of link-obstacle intersection volume moves link away from obstacle. (b) Gradient of swept volume between consecutive link poses prevents collision between trajectory waypoints
• Figure 4: Obstacle Expansion widens the thinner dimension, thus altering the gradient direction. This helps the manipulator find a configuration for retracting around the shelf.
• Figure 5: Left-to-right: Obstacle clearance per guide, guide count per cost function type, guide count per obstacle expansion type, guide count normalizing gradients, guide count per gradient weight, and individual guide contributions to the ensemble.