Figure 02

Figure 02, developed by Figure AI, represents a pivotal advancement in humanoid robotics, unveiled in August 2024 as the company's second-generation model. Standing at 168 cm tall and weighing 70 kg, it features a sleek matte-black exterior with fully integrated cabling throughout its limbs, enhancing durability, reliability, and suitability for industrial environments. The robot's architecture emphasizes a ground-up redesign, incorporating custom electric motors in each joint, a 2.25 kWh battery pack integrated into the torso for improved balance and up to 5 hours of runtime, and dual NVIDIA RTX GPU-based modules providing approximately 3x the on-device AI inference compute compared to its predecessor, Figure 01. At the core of Figure 02's intelligence is the Helix Vision-Language-Action (VLA) model, a generalist AI system that unifies perception, language understanding, and control. Helix employs a dual-system architecture: System 2 (S2), a 7B-parameter open-source Vision-Language Model operating at 7-9 Hz for semantic reasoning from monocular images, robot state, and natural language prompts; and System 1 (S1), an 80M-parameter transformer running at 200 Hz for high-rate visuomotor control across a 35-DoF upper-body action space, including wrists, fingers, torso, and head. Trained end-to-end on ~500 hours of teleoperated data with auto-labeled natural language instructions, Helix enables zero-shot generalization to thousands of novel household and industrial objects, precise manipulation, multi-robot collaboration, and real-time adaptation without task-specific fine-tuning. All inference runs onboard the dual embedded GPUs in a model-parallel fashion, ensuring low-latency, closed-loop operation. Complementing Helix is Figure's partnership with OpenAI for speech-to-speech capabilities, allowing natural voice conversations via onboard microphones and speakers. The vision system leverages six RGB cameras for perception, obstacle avoidance, and hand-eye coordination. Hands feature 16 degrees of freedom (4th generation design with 10 fingers mimicking human structure), capable of lifting 25 kg payloads at speeds up to 3 km/h walking. Real-world deployment history underscores Figure 02's maturity. In an 11-month pilot at BMW Group Plant Spartanburg starting in late 2024, a fleet of Figure 02 robots autonomously loaded sheet-metal parts onto welding fixtures, achieving >99% placement accuracy within 5 mm tolerance and 37-second load times. Over 1,250 operational hours across 10-hour shifts, they handled 90,000+ parts, contributed to 30,000+ BMW X3 vehicles, and logged 1.2 million steps (~200 miles), with minimal hardware failures primarily in the forearm subsystem due to thermal and cabling challenges. This data informed Figure 03's redesign, leading to Figure 02's retirement in November 2025. The deployment validated adaptive locomotion, field calibration, and AI-driven precision in dynamic manufacturing settings, marking a first for humanoid robots in production lines. Figure 02's blend of hardware robustness and end-to-end AI positions it as a bridge to scalable commercial humanoids.