Differentiable Spatial Planning using Transformers

Devendra Singh Chaplot, Deepak Pathak, Jitendra Malik

Keywords: Planning, Spatial planning, Path planning, Navigation, Manipulation, Robotics

Abstract: We consider the problem of spatial path planning. In contrast to the classical solutions which optimize a new plan from scratch and assume access to the full map with ground truth obstacle locations, we learn a planner from the data in a differentiable manner that allows us to leverage statistical regularities from past data. We propose Spatial Planning Transformers (SPT), which given an obstacle map learns to generate actions by planning over long-range spatial dependencies, unlike prior data-driven planners that propagate information locally via convolutional structure in an iterative manner. In the setting where the ground truth map is not known to the agent, we leverage pre-trained SPTs to in an end-to-end framework that has the structure of mapper and planner built into it which allows seamless generalization to out-of-distribution maps and goals. SPTs outperform prior state-of-the-art across all the setups for both manipulation and navigation tasks, leading to an absolute improvement of 7-19%.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

One-sentence Summary: A differentiable spatial planning model designed for long-distance spatial reasoning using Transformers which allows end-to-end mapping and planning without access to ground-truth maps.

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:2112.01010/code)

Reviewed Version (pdf): https://openreview.net/references/pdf?id=1ca9iAsu1w