Learning-Guided Planning in Large Dynamic Action Spaces: Budgeted Tree Search for One-to-Many Mobile Charging
LP-BTS narrows thousands of changing charging choices into a small searched set before acting.
The paper tests one-to-many mobile charging, where each stop can serve multiple nearby sensors and the available stops change as sensors fail. Its planner combines a graph proposal policy, a learned value critic, and budgeted PUCT search over short futures. In the sealed 30-scenario evaluation, LP-BTS had the highest observed survival and alive-AUC, but its edge over the strongest engineered comparator was not statistically resolved. ArXiv · AI/CL/LG's note
The paper tests one-to-many mobile charging, where each stop can serve multiple nearby sensors and the available stops change as sensors fail. Its planner combines a graph proposal policy, a learned value critic, and budgeted PUCT search over short futures. In the sealed 30-scenario evaluation, LP-BTS had the highest observed survival and alive-AUC, but its edge over the strongest engineered comparator was not statistically resolved. ArXiv · AI/CL/LG's note
score 4