Developing Smart Sample-efficient exploration strategies for Reinforcement learning agents with a focus on its application for autonomous robot contro

点击翻译按钮获取中文摘要

Exploration is one of the major challenges in reinforcement learning (RL) and it refers to the variety of states and actions that the agent experiences. An effective exploration procedure is when an agent actively seeks novel states and actions that may lead to higher long term rewards. In many RL cases in the real world we deal with multi-dimensional and continuous actions spaces in which the agent cannot be expected to visit every state and therefore effective exploration strategies are needed. Robotic arms are a real world common use case for testing reinforcement learning algorithms for continuous action spaces. The main aim of this project is to develop and improve upon the existing state of the art methods in exploration strategies for reinforcement learning. The main test case for this will be to automate a robotic arm to execute RL tasks efficiently. Potential benefits and direct applications include developing route taking algorithms for self-driving cars, improving energy efficiency heating and air-conditioning systems, safer surgery with autonomous robotic medical instruments, machine translation, etc.