Julen Urain, PhD Candidate and researcher from the Technische Universität Darmstadt, has presented the paper “Composable Energy Policies for Reactive Motion Generation and Reinforcement Learning”, during the Robotics Science and Systems Conference which took place virtually from 12th – 16th of July 2021.
The paper introduces Composable Energy Policies (CEP), a novel framework for modular reactive motion generation, which computes the control action by optimisation over the product of a se of stochastic policies. This product of policies can provide a high probability to those actions which satisfy all the components and low probability to the others. Optimizing over the product of the policies helps to avoid the detrimental effect of conflicting behaviours between policies choosing an action that satisfies all the objectives.
This document also defines how CEP naturally adapts to the Reinforcement Learning problem allowing us to integrate, in a hierarchical fashion, any distributions as prior, from multimodal distributions to non-smooth distributions and learn a new policy given them.
The Robotics: Science and Systems Conference counted with the participation of international researchers working in all areas of robotics sector and presented a programme with a huge variety of single-track presentations, workshops, poster sessions, tutorials and scientific sessions.
The Technische Universität Darmstadt is one of Germany’s leading technical universities and its Intelligent Autonomous Systems Institute is one of the strongest robotic learning groups in Europe, contributing to the Sharework project by leading the development of learning algorithms and the study and formalisation of human gestures.