VER: Scaling On-Policy RL Leads to the Emergence of Navigation in Embodied Rearrangement.
Erik WijmansIrfan EssaDhruv BatraPublished in: CoRR (2022)
Keyphrases
- optimal policy
- reinforcement learning
- markov decision process
- action selection
- state space
- markov decision processes
- partially observable domains
- control policy
- multi agent
- model free
- control policies
- policy evaluation
- average reward
- policy gradient
- learning algorithm
- policy search
- average cost
- function approximation
- reinforcement learning problems
- actor critic
- embodied cognition
- state and action spaces
- markov decision problems
- learning agents
- state action
- policy iteration
- temporal difference
- information space
- long run
- sufficient conditions