Motion Planning Diffusion

Learning and Planning of Robot Motions with Diffusion Models

João Carvalho, An T. Le, Mark Baier, Dorothea Koert and Jan Peters

2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 

Abstract: Learning priors on trajectory distributions can help accelerate robot motion planning optimization. Given previously successful plans, learning trajectory generative models as priors for a new planning problem is highly desirable. Prior works propose several ways on utilizing this prior to bootstrapping the motion planning problem. Either sampling the prior for initializations or using the prior distribution in a maximum-a-posterior formulation for trajectory optimization. In this work, we propose learning diffusion models as priors. We then can sample directly from the posterior trajectory distribution conditioned on task goals, by leveraging the inverse denoising process of diffusion models. Furthermore, diffusion has been recently shown to effectively encode data multimodality in high-dimensional settings, which is particularly well-suited for large trajectory dataset. To demonstrate our method efficacy, we compare our proposed method - Motion Planning Diffusion - against several baselines in simulated planar robot and 7-dof robot arm manipulator environments. To assess the generalization capabilities of our method, we test it in environments with previously unseen obstacles. Our experiments show that diffusion models are strong priors to encode high-dimensional trajectory distributions of robot motions.

Experiment Demos

The following videos show examples of planning with Motion Planning Diffusion (MPD) in simulated environments and in the real-world Panda Shelf environment.

Obstacles in red were not present in the training data.

PointMass2D Dense

These images show the sampling of trajectories with MPD from different start and goal configurations.
Notice how the trajectories evolve from Gaussian noise to collision-free trajectories, avoiding obstacles not seen during training.
In orange are depicted collision-free trajectories, and in black are trajectories in collision. 

Panda Spheres

The following videos show planning results from the same start and goal configurations.
Notice the multimodality of trajectories (variability of movements) generated between the videos. 

Panda Shelf

The following videos show planning results from the same start and goal configurations.
Notice the multimodality of trajectories (variability of movements) generated between the videos. 

Panda Shelf Real

The following videos show planning results from the same start and goal configurations.
Apart from collision-free generation, the EE cost function enforces the EE orientation to remain constant along the trajectory.
Notice the multimodality of trajectories (variability of movements) generated between the videos.