Deep Dive Papers

Recommended Learning Path

For the best learning experience, we recommend reading the papers in order. Each paper builds upon concepts from the previous ones.

1. 1
   Waymax Deep Dive

   Core simulator architecture, data-driven simulation, metrics system, and evaluation framework for autonomous driving.

2. 2
   WOSAC Challenge Deep Dive

   Waymo Open Sim Agents Challenge evaluation framework, realism metrics, and winning strategies from 2023-2025.

3. 3
   JAX Scaling RL Deep Dive

   How to scale RL training across GPUs/TPUs using JAX primitives: jit, vmap, pmap, scan, and distributed PPO.

4. 4
   V-Max Framework Deep Dive

   Complete RL training pipeline on top of Waymax including ScenarioMax, observation design, and reward hierarchy.

5. 5
   BehaviorGPT Deep Dive

   State-of-the-art sim agent modeling with transformers, Next-Patch Prediction, and the 2024 WOSAC winner approach.

6. 6
   Sim-to-Real Gap Deep Dive

   Bridging virtual and physical worlds: perception, actuation, and behavioral gaps with neural rendering and world models.

7. 7
   Long-Tail Scenarios Deep Dive

   Safety-critical testing at scale: adversarial generation, scenario mining, and coverage metrics for AV validation.

8. 8
   Distributed Training Deep Dive

   Scaling RL to billions of steps: PureJaxRL, actor-learner architectures, and GPU-accelerated simulation infrastructure.