Continuously Learning, Adapting, and Improving: A Dual-Process Approach to Autonomous Driving

Authors: Jianbiao Mei, Yukai Ma, Xuemeng Yang, Licheng Wen, Xinyu Cai, Xin Li, Daocheng Fu, Bo Zhang, Pinlong Cai, Min Dou, Botian Shi, Liang He, Yong Liu, Yu Qiao

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Closed-loop testing in CARLA shows that Leap AD outperforms all methods relying solely on camera input, requiring 1-2 orders of magnitude less labeled data. Experiments also demonstrate that as the memory bank expands, the Heuristic Process with only 1.8B parameters can inherit the knowledge from a GPT-4 powered Analytic Process and achieve continuous performance improvement. Project page: https://pjlab-adg.github.io/Leap AD/.
Researcher Affiliation Collaboration 1 Zhejiang University 2 Shanghai Artificial Intelligence Laboratory 3 East China Normal University 4 Shanghai Jiao Tong University
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The abstract mentions a "Project page: https://pjlab-adg.github.io/Leap AD/", which is a project overview page, not a direct link to a source-code repository, nor is there an explicit statement about code release for the methodology.
Open Datasets Yes We construct the instruct-following datasets for supervised fine-tuning of our VLM by integrating Rank2Tell [58], Drive LM [36], and data collected within CARLA [59].
Dataset Splits No The paper discusses training data and evaluation on a test benchmark, but does not explicitly define or specify a separate validation dataset split with percentages or sample counts for model tuning during development.
Hardware Specification Yes The batch size is set to 16, and the model is trained for 5 epochs on 8 A100 GPUs, requiring about 26 hours.
Software Dependencies No The paper mentions specific models like "Qwen-VL-7B", "GPT-4", and "Qwen1.5-1.8B" and the "Open AI embedding model", and the CARLA simulator, but does not provide specific version numbers for underlying software environments (e.g., Python, PyTorch, CUDA, or CARLA version).
Experiment Setup Yes We utilize the Adam W optimizer [61] with β1 = 0.9 and β2 = 0.95, coupled with a cosine decay of the learning rate, initially set to 1e 5. The batch size is set to 16, and the model is trained for 5 epochs on 8 A100 GPUs, requiring about 26 hours.