TensorPlan and the Few Actions Lower Bound for Planning in MDPs under Linear Realizability of Optimal Value Functions.
Gellért WeiszCsaba SzepesváriAndrás GyörgyPublished in: ALT (2022)
Keyphrases
- lower bound
- stochastic domains
- markov decision processes
- optimal solution
- partially observable
- upper bound
- decision theoretic planning
- action sets
- worst case
- state space
- initial state
- macro actions
- competitive ratio
- action selection
- reinforcement learning
- dynamic programming
- planning problems
- decision theoretic
- reward function
- plan recognition
- average cost
- goal state
- optimal planning
- linear functions
- decision diagrams
- branch and bound
- classical planning
- planning under uncertainty
- closed form
- objective function
- branch and bound algorithm
- markov decision problems
- state and action spaces
- temporally extended
- optimal control
- optimal plans
- dec pomdps
- dynamical systems
- average reward
- external events
- domain independent
- optimal policy
- heuristic search
- plan execution
- regret bounds
- probabilistic planning
- constant factor
- finite horizon