End-to-End Egospheric Spatial Memory

Published at the International Conference on Learning Representations (ICLR) 2021
End-to-End Egospheric Spatial Memory (ESM) encodes the memory in an ego-sphere around the agent, enabling expressive 3D representations. ESM can be trained end-to-end via either imitation or reinforcement learning, and improves both training efficiency and final performance against other memory baselines on various visuomotor control tasks.
Project page: djl11.github.io/ESM/
Paper: arxiv.org/abs/1810.03237​​
Authors: Daniel Lenton, Stephen James, Ronald Clark, Andrew J. Davison, Dyson Robotics Labl, Imperial College London.
Contact: [e-mail]