Boston Dynamics uses reinforcement learning and supervised learning from human demonstrations to train its Atlas humanoid robot to perform autonomous industrial tasks such as part sequencing in automotive manufacturing.

Boston Dynamics trains Atlas using three parallel approaches: teleoperation via virtual reality headsets to generate demonstration data, reinforcement learning in simulation where the robot practices millions of virtual repetitions, and motion capture of human movements to teach physical behaviors. The company confirmed in 2025 that its primary application was part sequencing in automotive manufacturing at Hyundai facilities. As of January 2026, the product version of Atlas was unveiled at CES, with the entire 2026 production run committed to Hyundai and Google DeepMind.