TerraTransfer++: Demonstration-Free End-to-End Planner
TerraTransfer++: Demonstration-Free End-to-End Planner
TerraTransfer++ is a demonstration-free end-to-end planner for autonomous driving that uses self-play in vectorized simulators to generate diverse and reactive policies. The method pretrains a single policy through self-play, distills its rollouts into a trajectory teacher, and trains lightweight adapters on a frozen vision backbone to reproduce the latent space of the teacher. The approach achieves strong performance on benchmark tasks such as HUGSim and NAVSIM without requiring human demonstrations.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
end-to-end performance
59.4
View original evidence
itation-based end-to-end method on the photorealistic 3D Gaussian splatting closed-loop benchmark HUGSim, and a planner that reaches 59.4 EPDMS on NAVSIM navhard, above every published method including DriveZero-Scale (57.1), and 95.5 PDMS on navtest, without human demonstrations.
Open source S6end-to-end performance
95.5
View original evidence
itation-based end-to-end method on the photorealistic 3D Gaussian splatting closed-loop benchmark HUGSim, and a planner that reaches 59.4 EPDMS on NAVSIM navhard, above every published method including DriveZero-Scale (57.1), and 95.5 PDMS on navtest, without human demonstrations.
Open source S6end-to-end performance
57.1
View original evidence
itation-based end-to-end method on the photorealistic 3D Gaussian splatting closed-loop benchmark HUGSim, and a planner that reaches 59.4 EPDMS on NAVSIM navhard, above every published method including DriveZero-Scale (57.1), and 95.5 PDMS on navtest, without human demonstrations.
Open source S6
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
itation-based end-to-end method on the photorealistic 3D Gaussian splatting closed-loop benchmark HUGSim, and a planner that reaches 59.4 EPDMS on NAVSIM navhard, above every published method including DriveZero-Scale (57.1), and 95.5 PDMS on navtest, without human demonstrations.
Open source S6
Source：arXiv Robotics — research abstracts · arxiv.org