Abstract:

A key challenge in complex visuomotor control is learning abstract representations that are effective for specifying goals, planning, and generalization. To this end, we introduce universal planning networks (UPN). UPNs embed differentiable planning within a goal-directed policy. This planning computation unrolls a forward model in a latent space and infers an optimal action plan through gradient descent trajectory optimization. The plan-by-gradient-descent process and its underlying representations are learned end-to-end to directly optimize a supervised imitation learning objective. We find that the representations learned are not only effective for goal-directed visual imitation via gradient-based trajectory optimization, but can also provide a metric for specifying goals using images. The learned representations can be leveraged to specify distance-based rewards to reach new target states for model-free reinforcement learning, resulting in substantially more effective learning when solving new tasks described via image-based goals. We were able to achieve successful transfer of visuomotor planning strategies across robots with significantly different morphologies and actuation capabilities.

Explanation of the Architecture (Video with audio)

Goal Conditioned Visual Imitation

Imitation Learning Results with simulated 2D robots

Transfer Using Reinforcement Learning

Reacher Morphology Transfer Results

Ant Navigation Results

Transfer Results from Poking to Pushing

Long Horizon Ant Navigation (Parkour Like Environment)

Humanoid Navigation Around Obstacle Walls

Six and Seven Link Reacher Results