https://github.com/leggedrobotics/digbench
Benchmarks and map generation for the Terra environment.
https://github.com/leggedrobotics/digbench
Last synced: about 1 year ago
JSON representation
Benchmarks and map generation for the Terra environment.
- Host: GitHub
- URL: https://github.com/leggedrobotics/digbench
- Owner: leggedrobotics
- Created: 2023-06-15T14:04:58.000Z (about 3 years ago)
- Default Branch: main
- Last Pushed: 2024-03-21T03:05:18.000Z (over 2 years ago)
- Last Synced: 2025-04-06T20:46:32.435Z (about 1 year ago)
- Language: Python
- Homepage:
- Size: 6.41 MB
- Stars: 6
- Watchers: 17
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Digbench
Digbench provides essential tools for generating maps to train Terra agents. It leverages both procedurally generated environments and real-world building footprints.
## Available Map Types
- **Foundations**: Downloaded from OpenStreetMap and projected onto a grid map.
- **Trenches**: Procedurally generated trenches featuring 1, 2, or 3 axes, along with obstacles, no-dumping zones, and terminal dumping constraints.
## Generating Maps
### Step 1: Generate Foundation Maps
1. Download and generate the foundation maps:
```bash
python generate_foundations.py
```
If you prefer not to download the entire set of foundations, you can interrupt the script and rerun it. The script will automatically format the maps correctly.
### Step 2: Create Training Data
1. Generate procedural trenches, add constraints and obstacles, and reformat the maps for Terra use:
```bash
python create_train_data.py
```
### Step 3: Verify Map Generation
1. Ensure the maps are correctly generated by running:
```bash
DATASET_PATH="/terra/digbench/data/train/" DATASET_SIZE= python -m terra.viz.play
```
Replace `` with the actual path to your Terra installation and `` with the desired dataset size.
sql