Summary of Module-wise Adaptive Adversarial Training For End-to-end Autonomous Driving, by Tianyuan Zhang et al.
Module-wise Adaptive Adversarial Training for End-to-end Autonomous Driving
by Tianyuan Zhang, Lu Wang, Jiaqi Kang, Xinwei Zhang, Siyuan Liang, Yuwei Chen, Aishan Liu, Xianglong Liu
First submitted to arxiv on: 11 Sep 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary A novel approach to enhancing the robustness of end-to-end autonomous driving (AD) models is proposed, which integrates perception, prediction, and planning stages. The method, called Module-wise Adaptive Adversarial Training (MA2T), combines two techniques: Module-wise Noise Injection and Dynamic Weight Accumulation Adaptation. These modules work together to adaptively learn and adjust the loss weights of each module based on their contributions for better balance and robust training. Experiments conducted on the nuScenes dataset show that MA2T outperforms other baselines by large margins (+5-10%) under both white-box and black-box attacks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about making self-driving cars safer by teaching them to resist fake information. Researchers have made big progress in this area, but they still need to improve how their models work together. The new method, called MA2T, does just that by adding noise to the signals that different parts of the model use. This helps the model learn to make better decisions and ignore any fake information it gets. Tests show that this approach makes the self-driving cars much more reliable. |
