Teaching a Robot Three-Point Turn (www.atirobotics.ai)

A team successfully developed a three-point turn maneuver for the Ati Robotics Sherpa 10K, a powerful robot designed for heavy hauling but challenged by tight spaces. Inspired by Andrej Karpathy’s autoresearch methodology, the project combined AI-driven iterations with real-world data from manual three-point turns to create a reliable, planner-callable solution. The core task was to allow the Sherpa 10K to perform turns in aisles barely wider than its footprint, addressing customer feedback regarding maneuverability. Significantly, this effort demonstrated the potential for AI in practical robotics design by utilizing a simple iterative process and real empirical data. A bespoke simulator served as a crucial tool for testing various design strategies, revealing valuable insights about the vehicle's kinematics and the constraints of turning in tight spaces. Ultimately, the project highlighted that success lay not just in AI generating specifications but in pruning unnecessary complexity and validating designs through simulation, thus paving the way for future enhancements in robotic navigation and maneuverability.