🦼🤖 Sneak peek: Beyond our own ATV autonomy results, we have also been pushing towards:
1. More challenging scenarios (Nighttime)
2. More robot platforms (quadruped & urban wheelchair) (5/5)

🗄️ What powers all this: multi-modal data, which we provide to everyone through TartanDrive 2.0, a large-scale off-road driving dataset for self-supervised learning tasks. Check out the data here: theairlab.org/TartanDrive2/ (4/5)

📍Since there is no prior map or GPS, our system relies on SLAM to track position and build a local understanding of the environment in real-time, enabling consistent perception across diverse terrains.

See the voxel grid from our SLAM point cloud colored by VFM features (3/5)

🧠 3 key principles: Self-supervised, Multi-modality, and Uncertainty Awareness.

Our system uses multi-modality (LiDAR+camera) data and learns by itself where to go (traversability) while balancing risk (uncertainty) and performance with inverse reinforcement learning. (2/5)