Hybrid Online POMDP Planning and Deep Reinforcement Learning for Safer Self-Driving Cars.
Florian PusseMatthias KluschPublished in: IV (2019)
Keyphrases
- reinforcement learning
- partially observable
- partially observable markov decision processes
- state space
- partially observable markov decision process
- planning problems
- sequential decision making problems
- continuous state
- belief space
- markov decision processes
- action selection
- online learning
- optimal policy
- partial observability
- hidden state
- function approximation
- learning algorithm
- policy search
- reinforcement learning algorithms
- model free
- stochastic domains
- belief state
- dynamical systems
- markov decision process
- planning under uncertainty
- complex domains
- markov decision problems
- partially observable stochastic domains
- multi agent
- temporal difference
- policy evaluation
- blocks world
- dynamic programming
- state and action spaces
- single agent
- model free reinforcement learning
- decision theoretic
- partially observable domains
- finite state
- machine learning
- partially observable stochastic games
- dec pomdps
- infinite horizon
- ai planning
- real time
- planning process