Reinforcement Learning for Sparse-Reward Object-Interaction Tasks in a First-person Simulated 3D Environment.
Wilka CarvalhoAnthony LiangKimin LeeSungryull SohnHonglak LeeRichard L. LewisSatinder SinghPublished in: IJCAI (2021)
Keyphrases
- reinforcement learning
- real robot
- multi agent environments
- agent learns
- sensory inputs
- object manipulation
- initially unknown
- transfer learning
- simulated robot
- mobile robot
- human computer interaction
- robotic systems
- learning agent
- test bed
- function approximation
- d objects
- optimal policy
- agent environment
- markov decision processes
- partially observable environments
- tabula rasa
- real time
- machine learning
- physical objects
- autonomous robots
- human users
- human operators
- state space
- user interaction
- multi agent
- agent receives
- complex domains
- physical world
- multiple objects
- high dimensional
- physical space
- real world environments
- exploration strategy
- simulation model
- dynamic programming
- autonomous agents
- learning tasks
- external world
- previously learned
- partially observable
- reinforcement learning algorithms