Bittensor Subnet SN80 Gets Decentralized Data-Collection Platform Shift

OpenRoboto has launched Shift, a decentralized data-collection platform on the Bittensor subnet SN80. The goal is to gather real-world video footage of environments like factories and kitchens to train vision-language-action robotics models.

Shift works within the Bittensor ecosystem, which already runs model-improvement competitions where miners refine open-weight robotics models and compete against randomized LIBERO-Pro benchmarks.

The subnet has been running these competitions since July 29, 2026, with a 70/20/10 split in emission distribution among top performers. As of early September 2026, 85% of emissions were flowing toward physical-robot competitions.

Shift adds a new dimension to this setup by creating a parallel track for data collection. Contributors are rewarded for submitting egocentric video footage that captures how humans interact with objects and navigate spaces.

The platform feeds the training pipeline for vision-language-action models built on foundations like π0.5 and LingBot-VLA 2.0. Axis Robotics has contributed over 3 million multimodal trajectories to an Open Data Pool, which combines visual, spatial, and action data.