CVJun 11, 2024

Hydra-MDP: End-to-end Multimodal Planning with Multi-target Hydra-Distillation

arXiv:2406.06978v4202 citationsHas Code
Originality Incremental advance
AI Analysis

This addresses planning for autonomous driving, offering a novel approach but appears incremental as it builds on existing distillation and multi-head decoder techniques.

The paper tackles multimodal planning in driving by using a teacher-student model with knowledge distillation from human and rule-based teachers, achieving first place in the Navsim challenge with improved generalization across environments.

We propose Hydra-MDP, a novel paradigm employing multiple teachers in a teacher-student model. This approach uses knowledge distillation from both human and rule-based teachers to train the student model, which features a multi-head decoder to learn diverse trajectory candidates tailored to various evaluation metrics. With the knowledge of rule-based teachers, Hydra-MDP learns how the environment influences the planning in an end-to-end manner instead of resorting to non-differentiable post-processing. This method achieves the $1^{st}$ place in the Navsim challenge, demonstrating significant improvements in generalization across diverse driving environments and conditions. More details by visiting \url{https://github.com/NVlabs/Hydra-MDP}.

Code Implementations3 repos
Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes