You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
-[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)
48
45
49
46
## Highlights
50
47
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.
52
49
-**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.
53
50
-**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.
54
51
-**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
68
65
> **Metric notes:**
69
66
> -**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`.
70
67
> -**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.
72
69
>
73
70
> **Training notes:**
74
71
> -**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
89
86
- 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.
90
87
- 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.
### Qualitative Results — Closed-Loop Simulation on nuPlan
93
90
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.
95
92
96
93
<divalign="center">
97
94
<table>
@@ -137,7 +134,6 @@ WorldEngine consists of two tightly coupled subsystems:
0 commit comments