Solana-Powered BitRobot Unleashes 2,000 Hours of Urban Navigation Data on Embodied AI
BitRobot, a Solana (SOL)-powered robotics network, is open-sourcing 2,000 hours of urban navigation data to advance embodied AI. The dataset dwarfs previous public datasets, which were limited to just 60 hours.
The project started as FrodoBots, with sidewalk robots operated remotely by gamers producing valuable navigation data during scavenger hunt-style missions. Teams from DeepMind, Meta, and UC Berkeley have already utilized the dataset to train navigation models, demonstrating its value to the AI community.
Jonathan Victor, president of BitRobot, argues that real-world data is a key bottleneck for robotics, surpassing even compute power and model design. To tackle this, BitRobot has created a network of task-specific subnets, each designed to collect unique types of data, from urban navigation to dexterity tasks.
BitRobot's flagship tool is the RoboCap, a $1,000 wearable device equipped with six cameras that capture first-person video and hand movements. Contributors earn rewards based on the novelty and utility of their data, determined by a scoring system that prioritizes unique scenarios over repetitive tasks.