Distillation of Reasoning Models for Efficient Route Planning
Distill large reasoning models into efficient student planners via chain-of-thought trace generation to enable fast mobile route planning.
Source paper
Zhiheng Song, Jingshuai Zhang, Chuan Qin, Chao Wang, Chao Chen, Longfei Xu, Kaikui Liu, Xiangxiang Chu, Hengshu Zhu
Generated by AI. The paper’s authors did not write these.
Distill large reasoning models into efficient student planners via chain-of-thought trace generation to enable fast mobile route planning.
Enforce strict user preferences in route planning by parsing intents into formal logic specifications evaluated by a deterministic constraint solver.
Dynamic balance agent planning efficiency and robustness by switching from execution-based plans to ReAct loops only when errors occur.
This paper introduces MobilityBench, a scalable and reproducible benchmark designed to evaluate Large Language Model (LLM) based agents on real-world route-planning tasks. The authors constructed the dataset from over 100,000 anonymized real-user queries from Amap, established a deterministic API-replay sandbox to eliminate environmental variance (e.g., changing traffic), and proposed a multi-dimensional evaluation protocol to assess capabilities like instruction understanding, tool use, and efficiency.
The core methodology involves an 'episode-centric' formulation where agents interact with a deterministic sandbox that replays cached API responses (POIs, routing, traffic) corresponding to historical real-world states. Agents are evaluated using a taxonomy of 11 intent scenarios ranging from simple POI retrieval to complex preference-constrained planning, with metrics focusing on execution correctness (Delivery Rate, Final Pass Rate) and process quality (efficiency, tool schema compliance).