… To tackle this challenge, we propose Learning from Tree Search for Driving (LeTS-Drive),
which integrates planning and learning in a closed loop. The algorithm comprises the …

… -Carlo Tree Search method, named as KBTree, for motion planning in autonomous driving.
… ) as our feature extractor to improve the learning performance. To the best of our knowledge…

… It bridges forward belief tree search with heuristics produced by an offline backward solver.
… such as driving in an urban crowd. In this paper, we propose a principled approach to learn …

… of a Monte Carlo Tree Search algorithm with a deep-learning heuristic. We learn a fast …
Lets-drive: Driving in a crowd by learning from tree search. CoRR, abs/1905.12197, 2019. …

… Most recently, LeTS-Drive [20] used online belief-tree search to learn a value and policy
function for autonomous driving in a crowded space. Although this approach models intentions …

… LeTS-Drive [25] proposed an algorithm using an online belief-tree search. They utilize
Convolutional … Driving in crowds can be regarded as a sequential decision making problem in …

… LeTS-Drive [31] learns driving policies from a sparsely sampled tree search-based
planner, then uses this learned policy to guide real-time vehicle control. Although it tries to …

… ) as feedback to learn an end-to-end macroaction generator. Another closed-loop work, called
LeTSDrive [17], learns driving policies from a sparsely sampled tree search-based planner…

… in a search algorithm, such as Monte Carlo Tree Search (… to guide the search, vastly improving
search efficiency. Experiments … Lets-drive: Driving in a crowd by learning from tree search. …

… The planner then outputs optimal driving policies from the search tree and re- … to control the
driving speed. Later, LeTS-Drive [13] extended the work by integrating POMDP planning with …