added: seal

_posts/2024-09-19-SEAL.md

@@ -0,0 +1,90 @@
+---
+layout: post
+title:  "SEAL: Towards Safe Autonomous Driving via Skill-Enabled Adversary Learning for Closed-Loop Scenario Generation"
+date:   2024-09-19 10:00:00
+categories: research
+description: "Closed-Loop Scenario Generation"
+author: "Benjamin Stoler"
+published: true
+sidebar:  false
+hide_hero: true
+permalink: /seal/
+image: /assets/imgs/posts/2024-09-19-SEAL/main.png
+link-new-tab: true
+---
+<hr>
+
+[Benjamin Stoler](https://benstoler.com), [Ingrid Navarro](https://navars.xyz), Jonathan Francis and [Jean Oh](https://cmubig.github.io/team/jean_oh/) 
+
+<a class="button" itemprop="github" href="https://github.com/cmubig/SEAL" target="_blank">
+  <i class="fab fa-github fa-lg"></i>
+</a>
+<a class="button" itemprop="paper" href="https://arxiv.org/abs/2409.10320" target="_blank">
+  <i class="fas fa-file fa-lg"></i>    
+</a>
+
+# Abstract
+
+Verification and validation of autonomous driving (AD) systems and components is of increasing importance, as such technology increases in real-world prevalence. Safety-critical scenario generation is a key approach to robustify AD policies through closed-loop training. However, existing approaches for scenario generation rely on simplistic objectives, resulting in overly-aggressive or non-reactive adversarial behaviors. To generate diverse adversarial yet realistic scenarios, we propose SEAL, a scenario perturbation approach which leverages learned scoring functions and adversarial, human-like skills. SEAL-perturbed scenarios are more realistic than SOTA baselines, leading to improved ego task success across real-world, in-distribution, and out-of-distribution scenarios, of more than 20%. 
+<hr>
+
+# Method
+
+### Overview
+
+todo
+
+### Motivation
+
+#### Scenario Realism
+
+todo
+
+<p align="center">
+  <img width="1280" src="/assets/imgs/posts/2024-09-19-SEAL/example.png" alt="Realism">
+</p>
+
+#### Evaluation Fairness
+
+todo
+
+<p align="center">
+  <img width="1280" src="/assets/imgs/posts/2024-09-19-SEAL/example.png" alt="Fairness">
+</p>
+
+<hr>
+
+# Method
+
+### Learned Score Function
+
+todo
+
+### Adversarial Skill Learning
+
+<p align="center">
+  <img width="500" src="/assets/imgs/posts/2024-09-19-SEAL/skill_space.png" alt="Skills">
+</p>
+
+# Results
+
+#### Ego Policy Training
+
+todo
+
+#### SEAL-generated Scenario Quality 
+
+todo
+
+<hr>
+
+# BibTeX
+
+```
+@article{stoler2024seal,
+  title={SEAL: Towards Safe Autonomous Driving via Skill-Enabled Adversary Learning for Closed-Loop Scenario Generation},
+  author={Stoler, Benjamin and Navarro, Ingrid and Francis, Jonathan and Oh, Jean},
+  journal={arXiv preprint arXiv:2409.10320},
+  year={2024}
+}
+```