Voronoi Progressive Widening: Efficient Online Solvers for Continuous State, Action, and Observation POMDPs
About
This paper introduces Voronoi Progressive Widening (VPW), a generalization of Voronoi optimistic optimization (VOO) and action progressive widening to partially observable Markov decision processes (POMDPs). Tree search algorithms can use VPW to effectively handle continuous or hybrid action spaces by efficiently balancing local and global action searching. This paper proposes two VPW-based algorithms and analyzes them from theoretical and simulation perspectives. Voronoi Optimistic Weighted Sparse Sampling (VOWSS) is a theoretical tool that justifies VPW-based online solvers, and it is the first algorithm with global convergence guarantees for continuous state, action, and observation POMDPs. Voronoi Optimistic Monte Carlo Planning with Observation Weighting (VOMCPOW) is a versatile and efficient algorithm that consistently outperforms state-of-the-art POMDP algorithms in several simulation experiments.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Control Task | Lunar Lander (test) | Average Reward60.94 | 31 | |
| Planning | 3D-Continuous Light-Dark | Mean Return4.8 | 30 | |
| POMDP Navigation | 4D-Continuous Light-Dark | Mean Return3.04 | 30 | |
| POMDP Planning | 2D-Continuous Light-Dark (test) | Mean Return6.05 | 30 | |
| Reinforcement Learning | Lunar Lander POMDP | Performance Score56.09 | 30 | |
| Hill Car POMDP | Hill Car POMDP | Mean Return78.37 | 30 | |
| Two-Agent 2D-Continuous Light-Dark Navigation | Two-Agent 2D-Continuous Light-Dark | Mean Performance2.57 | 30 | |
| Continuous Control | Mountain Car POMDP | Mean Performance25.62 | 30 | |
| Mountain Car | Mountain Car | Mean Return24.42 | 20 | |
| Reinforcement Learning | Hill Car MDP | Performance-59.66 | 20 |