Applying Neural Monte Carlo Tree Search to Unsignalized Multi-intersection Scheduling for Autonomous Vehicles
원문 초록
Dynamic scheduling of access to shared resources by autonomous systems is a challenging problem, characterized as being NP-hard. The complexity of this task leads to a combinatorial explosion of possibilities in highly dynamic systems where arriving requests must be continuously scheduled subject to strong safety and time constraints. An example of such a system is an unsignalized intersection, where automated vehicles' access to potential conflict zones must be dynamically scheduled. In this paper, we apply Neural Monte Carlo Tree Search (NMCTS) to the challenging task of scheduling platoons of vehicles crossing unsignalized intersections. Crucially, we introduce a transformation model that maps successive sequences of potentially conflicting road-space reservation requests from platoons of vehicles into a series of board-game-like problems and use NMCTS to search for solutions representing optimal road-space allocation schedules in the context of past allocations. To optimize search, we incorporate a prioritized re-sampling method with parallel NMCTS (PNMCTS) to improve the quality of training data. To optimize training, a curriculum learning strategy is used to train the agent to schedule progressively more complex boards culminating in overlapping boards that represent busy intersections. In a busy single four-way unsignalized intersection simulation, PNMCTS solved 95\% of unseen scenarios, reducing crossing time by 43\% in light and 52\% in heavy traffic versus first-in, first-out control. In a 3x3 multi-intersection network, the proposed method maintained free-flow in light traffic when all intersections are under control of PNMCTS and outperformed state-of-the-art RL-based traffic-light controllers in average travel time by 74.5\% and total throughput by 16\% in heavy traffic.
함께 살펴볼 논문
공통 태그를 근거로 연결했습니다. 실제 인용 관계나 논문의 우수성을 뜻하지 않습니다.
제약 만족 문제에서 형식화자로서 언어 모델에 대한 현실 점검
A Reality Check of Language Models as Formalizers on Constraint Satisfaction Problems
최근 연구는 기호적 추론(symbolic reasoning) 문제에서 대규모 언어 모델(LLM)을 end-to-end 솔버로 쓰기보다 형식화자(formalizer)로 사용할 때 더 우수한 성능을 보인다고 한다. 문제 설명이 주어지면 LLM은 외부 솔버를 통해 해법을 도출하는 형식 프로그램(formal program)을 생성한다. 우리는 4개 벤치마크, 6개 LLM, 2종류의 형식 언어(formal language)에 걸쳐 실제 제약 만족 문제에서 LLM의 형식화 역량을 체계적으로 조사한다. LLM-as-formalizer가 문제를 결코 사소하게 만들지는 않지만, 검증 가능성(verifiability)과 해석 가능성(interpretability)에도 불구하고 24개 모델-데이터셋 조합 중 15개에서 LLM-as-solver보다 성능이 낮음을 보인다. 형식화 공간이 탐색 공간(search space)보다 수 orders of magnitude 작음에도, 스케일링 분석 결과 LLM-as-formalizer는 LLM-as-solver와 마찬가지로 문제 복잡도가 증가할수록 급격히 저하된다. 이 한계를 더 잘 이해하기 위해, 때로는 하드코딩된 해법(hard-coded solution)으로 이어지는 과도한 솔버 유사 추론 토큰(solver-like reasoning token)을 관찰하였으며, 이는 LLM 기반 형식화 개선을 위한 핵심 과제임을 부각한다.
TIP-Search: Time-Predictable Inference Scheduling for Market Prediction under Uncertain Load
Optimal Sensor Scheduling and Selection for Continuous-Discrete Kalman Filtering with Auxiliary Dynamics
SystolicAttention: Fusing FlashAttention within a Single Systolic Array
A Survey on Vision-Language-Action Models for Autonomous Driving
Where, What, Why: Towards Explainable Driver Attention Prediction
토론
이 논문과 연결된 포럼 글이 아직 없습니다.