Summary of Cooperative Advisory Residual Policies For Congestion Mitigation, by Aamir Hasan et al.
Cooperative Advisory Residual Policies for Congestion Mitigation
by Aamir Hasan, Neeloy Chakraborty, Haonan Chen, Jung-Hoon Cho, Cathy Wu, Katherine Driggs-Campbell
First submitted to arxiv on: 30 Jun 2024
Categories
- Main: Machine Learning (cs.LG)
- 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 In this paper, researchers develop learned residual policies for cooperative advisory systems that can be used with a single vehicle featuring a human driver. These policies advise drivers to take actions that mitigate traffic congestion while accounting for diverse driver behaviors and reactions to instructions. The team introduces an improved reward function that addresses congestion mitigation and driver attitudes, and uses a variational autoencoder to personalize the policies based on inferred driver traits. The models are trained in simulation using a novel instruction adherence driver model and evaluated through both simulation tests and a user study (N=16). Results show up to 20% and 40% improvement in congestion mitigation compared to baselines, with personalized policies that are human-compatible. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Autonomous vehicles can help reduce traffic congestion by making simple changes. However, current approaches have limitations since they require precise control over many vehicles and expensive sensors. This paper proposes a new way for a single vehicle with a human driver to make decisions that improve traffic flow while considering different driver behaviors. The team creates policies that advise drivers on how to behave in ways that reduce congestion. They also develop a reward function that takes into account both congestion reduction and driver reactions. The results show that these approaches can successfully reduce congestion and adapt to different driver behaviors, with significant improvements. |
Keywords
* Artificial intelligence * Variational autoencoder
