Project Page | Video
This repo contains the implementation of our paper:
DRIVE: A Unified Framework for Empowering Personalized Autonomous Driving with Humanoid Robots
Note: The source code will be open-sourced upon paper acceptance.
The key hyperparameters and their values used in our method are summarized below:
| Hyperparameter | Value |
|---|---|
| Batch size | 2048 |
| Minibatch size | 64 |
| Learning rate (α) | 0.0001 |
| GAE parameter (λ) | 0.95 |
| Discount factor (γ) | 0.99 |
| Clipping factor (ε) | 0.2 |
| Number of reward models (K) | 3 |
| Number of Metrics (M) | 2 (Lateral Error & Speed Tracking Error) |
| Reward model initial weight (β₀) | 0.05 |
| Weight decay rate (ρ) | 0.001 |
| Smoothing factor (α) | 0.9 |
| Inertial coefficients (M₀) | [1,1,1,1,1,1] |
| Damping coefficients (D₀) | [0.1,0.1,0.1,5.5,5.5,5.5] |
| Stiffness coefficients (K₀) | [500,500,500,500,500,500] |
| Control period (Δt) | 1/100 Hz |
| PPO decision period | 1/15 Hz |
| Steering radius (R) | 0.23 m |
| Optimizer | Adam for all networks |