Pixel to policy: DQN Encoders for within & cross-game reinforcement learning.
Ashrya AgrawalPriyanshi ShahSourabh PrakashPublished in: CoRR (2023)
Keyphrases
- reinforcement learning
- optimal policy
- agent learns
- policy search
- markov games
- markov decision process
- action selection
- markov decision processes
- partially observable environments
- function approximators
- decision problems
- reinforcement learning algorithms
- function approximation
- policy evaluation
- markov decision problems
- reinforcement learning problems
- game theoretic
- state space
- video games
- reward function
- multiagent reinforcement learning
- policy gradient
- control policies
- actor critic
- approximate dynamic programming
- model free
- state and action spaces
- partially observable
- control policy
- game design
- learning algorithm
- action space
- dynamic programming
- video compression
- policy iteration
- inverse reinforcement learning
- optimal strategy
- game play
- educational games
- rl algorithms
- game playing
- game theory
- computer games
- continuous state spaces
- monte carlo tree search
- partially observable domains
- transition model
- policy gradient methods
- state action
- stochastic games
- learning agent
- partially observable markov decision processes
- temporal difference
- long run
- learning process
- monte carlo
- multi agent
- continuous state
- agent receives
- virtual world
- pixel values
- nash equilibrium
- serious games
- pixel level
- average reward
- imperfect information