Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
SAH-Drive: A Scenario-Aware Hybrid Planner for Closed-Loop Vehicle Trajectory Generation
Authors: Yuqi Fan, Zhiyong Cui, Zhenning Li, Yilong Ren, Haiyang Yu
ICML 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The experimental results show that the proposed method significantly improves the generalization capability of the planning system, achieving state-of-the-art performance in inter Plan, while maintaining computational efficiency without incurring substantial additional runtime. |
| Researcher Affiliation | Academia | 1School of Transportation Science and Engineering, Beihang University, Beijing, China 2State Key Laboratory of Intelligent Transportation System, Beihang University, Beijing, China 3School of Software Engineering, Beihang University, Beijing, China 4State Key Laboratory of Internet of Things for Smart City, University of Macau, Macao, China. Correspondence to: Zhiyong Cui <EMAIL>. |
| Pseudocode | Yes | Algorithm 1 Planner Selection Using Dual-Timescale Decision Neuron |
| Open Source Code | Yes | The code for this paper is available at https://github.com/richie-live/SAH-Drive. |
| Open Datasets | Yes | In this section, SAH-Drive is assessed within the nu Plan benchmark (Karnchanachari et al., 2024), a well-recognized framework that incorporates estimated perception data for vehicles, pedestrians, lanes, and traffic signs. ... Training dataset: The nu Plan Mini dataset is a compact version of the full nu Plan dataset, designed for efficient experimentation in autonomous driving. ... Validiation dataset: We use Val14 (Dauner et al., 2023) and Test14-Random (Cheng et al., 2024b) to evaluate the planner s performance in regular scenarios and inter Plan (Hallgarten et al., 2024) and Test14-Hard (Cheng et al., 2024b) to assess its performance in long-tail scenarios. |
| Dataset Splits | Yes | Validiation dataset: We use Val14 (Dauner et al., 2023) and Test14-Random (Cheng et al., 2024b) to evaluate the planner s performance in regular scenarios and inter Plan (Hallgarten et al., 2024) and Test14-Hard (Cheng et al., 2024b) to assess its performance in long-tail scenarios. ... Table 2: Comparison with SOTA planners on different splits of nu Plan. Including inter Plan (long tail), Val14 (R), Val14 (NR), Test14-Random (R), Test14-Random (NR), Test14-Hard (R), Test14-Hard (NR). |
| Hardware Specification | Yes | We conducted a runtime analysis of PDM-Closed and SAH-Drive on a computer with an i7-14700KF CPU and an RTX 4080S GPU. |
| Software Dependencies | No | The paper does not explicitly list specific software dependencies with version numbers (e.g., Python, PyTorch, CUDA versions). |
| Experiment Setup | No | The paper mentions mechanisms that use parameters and thresholds (e.g., "threshold τ" for proposal regulator, "s1 and s2" for score-based switching rule, "ne consecutive runs" and "np iterations" for scenario-based switching rule, and parameters like A+, A-, τ+, τ- for STDP-based decision neuron), but it does not provide the concrete numerical values for these hyperparameters or system-level training settings in the main text. For example, Equation (3) describes the exponential weighting mechanism for trajectory fusion, stating "The parameter α controls the sensitivity of the weighting to the trajectory scores" but does not specify a value for α. |