Skip to content

Commit 48eac94

Browse files
committed
update docs
1 parent 9d7b7a6 commit 48eac94

2 files changed

Lines changed: 28 additions & 33 deletions

File tree

README.md

Lines changed: 24 additions & 28 deletions
Original file line numberDiff line numberDiff line change
@@ -23,32 +23,29 @@
2323
> Joint effort by OpenDriveLab at The University of Hong Kong, Huawei Inc. and Shanghai Innovation Institute (SII).
2424
2525
## Table of Contents
26-
27-
- [Towards the Era of Post-Training for Physical AI](#towards-the-era-of-post-training-for-physical-ai)
28-
- [Table of Contents](#table-of-contents)
29-
- [Highlights](#highlights)
30-
- [News](#news)
31-
- [Benchmark](#benchmark)
32-
- [Qualitative Results — Closed-Loop Simulation (nuPlan)](#qualitative-results--closed-loop-simulation-nuplan)
33-
- [On-Road Deployment — Night Urban Driving](#on-road-deployment--night-urban-driving)
34-
- [System Architecture](#system-architecture)
35-
- [Roadmap](#roadmap)
36-
- [Getting Started](#getting-started)
37-
- [Documentation Overview](#documentation-overview)
38-
- [Installation](#installation)
39-
- [Quick Test](#quick-test)
40-
- [Deep Dive by Module](#deep-dive-by-module)
41-
- [SimEngine - Photorealistic Closed-Loop Simulation](#simengine---photorealistic-closed-loop-simulation)
42-
- [AlgEngine - End-to-End Model Training \& Fine-Tuning](#algengine---end-to-end-model-training--fine-tuning)
43-
- [RenderEngine - Multi-Traversal Scene Reconstruction](#renderengine---multi-traversal-scene-reconstruction)
44-
- [Citation](#citation)
45-
- [Contributing](#contributing)
46-
- [License](#license)
47-
- [Related Resources](#related-resources)
26+
- [Highlights](#highlights)
27+
- [News](#news)
28+
- [Benchmark](#benchmark)
29+
- [Qualitative Results — Closed-Loop Simulation on nuPlan](#qualitative-results--closed-loop-simulation-on-nuplan)
30+
- [On-Road Deployment — Night Urban Driving](#on-road-deployment--night-urban-driving)
31+
- [System Architecture](#system-architecture)
32+
- [Roadmap](#roadmap)
33+
- [Getting Started](#getting-started)
34+
- [Documentation Overview](#documentation-overview)
35+
- [Installation](#installation)
36+
- [Quick Test](#quick-test)
37+
- [Deep Dive by Module](#deep-dive-by-module)
38+
- [SimEngine - Photorealistic Closed-Loop Simulation](#simengine---photorealistic-closed-loop-simulation)
39+
- [AlgEngine - End-to-End Model Training \& Fine-Tuning](#algengine---end-to-end-model-training--fine-tuning)
40+
- [Scene Reconstruction - 3D Gaussian Splatting-based method, MTGS](#scene-reconstruction---3d-gaussian-splatting-based-method-mtgs)
41+
- [Citation](#citation)
42+
- [Contributing](#contributing)
43+
- [License](#license)
44+
- [Related Resources](#related-resources)
4845

4946
## Highlights
5047

51-
- **WorldEngine** is the first post-training framework for Physical AI that systematically addresses the long-tail safety-critical data scarcity problem in autonomous driving.
48+
- **WorldEngine** is a post-training framework for Physical AI that systematically addresses the long-tail safety-critical data scarcity problem in autonomous driving.
5249
- **Data-driven long-tail discovery**: Failure-prone scenarios are automatically identified from real-world driving logs by the pre-trained agent itself — no manual design, no synthetic perturbations.
5350
- **Photorealistic interactive simulation** via 3D Gaussian Splatting (3DGS): Each discovered scenario is reconstructed into a fully controllable, real-time-renderable simulation environment with independent dynamic agent manipulation.
5451
- **Behavior-driven scenario generation**: Leverages Behavior World Model (BWM) to generalize and synthesize diverse traffic variations from existing long-tail scenarios, expanding sparse safety-critical events into a dense, learnable distribution.
@@ -68,7 +65,7 @@ We compare different post-training paradigms on the nuPlan dataset, evaluating o
6865
> **Metric notes:**
6966
> - **Open-loop PDMS** is aligned with [NAVSIM v1.1](https://github.com/autonomousvision/navsim) PDM Score. *Common* denotes the standard `navtest` split; *Rare* denotes the `navtest_failures` subset — failure-prone rare-case scenarios extracted from `navtest`.
7067
> - **Closed-loop Success Rate** is defined as the fraction of simulated driving episodes completed without collision or off-road failure.
71-
> - **Closed-loop PDMS*** is the PDM Score obtained via SimEngine closed-loop testing, where the planner interacts with reactive agents in simulation under real-time rendering.
68+
> - **Closed-loop PDMS*** is the PDM Score obtained via **SimEngine closed-loop testing**, where the planner interacts with reactive agents in simulation under real-time rendering.
7269
>
7370
> **Training notes:**
7471
> - **Rare logs** are failure-prone scenarios automatically extracted from `navtrain` by the pre-trained agent itself (see [Rare Case Extraction](docs/algengine_usage.md#rare-case-extraction)).
@@ -89,9 +86,9 @@ We compare different post-training paradigms on the nuPlan dataset, evaluating o
8986
- Post-training on **common logs** provides limited benefit and even degrades closed-loop performance (success rate drops from 73.61% to 64.58%), confirming that long-tail event discovery is essential.
9087
- The full **WorldEngine** pipeline achieves the best closed-loop performance (**88.89%** success rate, **70.12** PDMS*), a **+15.28%** absolute improvement in success rate over the base model.
9188

92-
### Qualitative Results — Closed-Loop Simulation (nuPlan)
89+
### Qualitative Results — Closed-Loop Simulation on nuPlan
9390

94-
Each pair shows the **Base model (FAIL)** vs **WorldEngine post-trained model (PASS)** on the same rare-case scenario. Left: front-camera rendering; Right: BEV trajectory visualization.
91+
Each pair shows the **Base model** vs **WorldEngine post-trained model** on the same rare-case scenario. Left: front-camera rendering; Right: BEV trajectory visualization.
9592

9693
<div align="center">
9794
<table>
@@ -137,7 +134,6 @@ WorldEngine consists of two tightly coupled subsystems:
137134
- [x] Hugging Face / ModelScope dataset
138135
- [ ] Open-source release (code, data, pre-trained models)
139136
- [ ] arXiv preprint
140-
- [ ] Demo webpage with interactive visualization
141137
- [ ] Behavior World Model integration
142138

143139

@@ -212,7 +208,7 @@ Learn how to train models, extract rare cases, and fine-tune:
212208

213209
**[AlgEngine Usage Guide](docs/algengine_usage.md)**
214210

215-
#### RenderEngine - Multi-Traversal Scene Reconstruction
211+
#### Scene Reconstruction - 3D Gaussian Splatting-based method, MTGS
216212

217213
WorldEngine's simulation environments are powered by 3D Gaussian Splatting (MTGS):
218214

docs/installation.md

Lines changed: 4 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -3,7 +3,7 @@
33
WorldEngine consists of two main subsystems that require **separate conda environments** due to different Python version requirements:
44

55
- **SimEngine** (`simengine` env, Python 3.9) - Closed-loop simulation with photorealistic rendering
6-
- **AlgEngine** (`algengine` env, Python 3.8) - End-to-end model training and evaluation
6+
- **AlgEngine** (`algengine` env, Python 3.9) - End-to-end model training and evaluation
77

88
> **Note:** RenderEngine uses the same environment as SimEngine since it's based on MTGS and shares dependencies.
99
@@ -29,7 +29,6 @@ WorldEngine consists of two main subsystems that require **separate conda enviro
2929

3030
- **OS:** Linux (Ubuntu 20.04/22.04 recommended)
3131
- **CUDA:** 11.8
32-
- **GCC:** 9.5 (for tiny-cuda-nn compilation)
3332
- **Conda/Miniconda:** Latest version
3433

3534
---
@@ -127,7 +126,7 @@ This environment is used for:
127126
#### 1. Create Conda Environment
128127

129128
```bash
130-
conda create --name algengine python=3.8 -y
129+
conda create --name algengine python=3.9 -y
131130
conda activate algengine
132131
```
133132

@@ -236,9 +235,9 @@ print(f'✓ CUDA available: {torch.cuda.is_available()}')
236235

237236
| Feature | SimEngine (`simengine`) | AlgEngine (`algengine`) |
238237
|---------|---------------------------|-------------------------|
239-
| **Python Version** | 3.9 | 3.8 |
238+
| **Python Version** | 3.9 | 3.9 |
240239
| **PyTorch** | 2.0.1+cu118 | 2.0.1+cu118 |
241-
| **Key Libs** | gsplat, tiny-cuda-nn, ray | MMCV, MMDet3D, MMCls |
240+
| **Key Libs** | gsplat, ray | MMCV, MMDet3D, MMCls |
242241
| **Use Cases** | Simulation, Rendering, BWM | Training, Evaluation |
243242
| **Disk Space** | ~10 GB | ~15 GB |
244243
| **Install Time** | ~30 min | ~45 min (MMCV build) |

0 commit comments

Comments
 (0)