latest

v0.2.0

cleaned up, integrated semantic labels with tree trunks (from wildsce…

v0.1.1

• The pre-trained models predict not only the terrain shape, but also friction values.
• New Terrain Encoder models are added: LidarBEV and BEVFusion. The architectures allow to use lidar scans as an additional sensor input for terrain properties prediction.

ICML 2024 DiffXYZ

ICML 2024 DiffXYZ

The package version accepted to the Differentiable Almost Everything Workshop: https://differentiable.xyz/

IROS-2024 submission

iros-2024

trajectory of the footprint in evaluation (fixed for rellis3d evaluat…

Checkpoints

Terrain Encoder weights trained on ROUGH and RELLIS-3D datasets.
The PyTorch checkpoint names have the following format: {model}_{dataset}_{robot}.pt