Figure 02 — Figure AI's second-generation humanoid robot — shipped its developer SDK in early 2026, opening what has been a closed research platform to third-party application development for the first time. For robotics engineers and AI developers evaluating humanoid platforms, the Figure 02 SDK represents a bet on the most physically capable commercial humanoid available, with BMWs production deployment providing the strongest industrial validation. This guide covers the Figure 02 SDK, the development workflow, and the application patterns that are emerging in early developer deployments.
Figure 02 Specifications
| Specification | Figure 02 | Notes |
|---|---|---|
| Height | 5'6" (168cm) | Designed to fit human workspaces without modification |
| Weight | 70kg | Comparable to average adult |
| Payload | 20kg | Sufficient for most manufacturing material handling |
| Battery life | 5+ hours continuous operation | Hot-swap battery capability for 24/7 operation |
| Hand DOF | 16 DOF per hand | Dexterous manipulation; individual finger control |
| Onboard compute | NVIDIA Orin + custom accelerator | Real-time vision processing and motor control |
| Connectivity | Wi-Fi 6E, 5G optional | Remote operation and model update capability |
Figure 02 SDK Architecture
Developer access requires Figure AI developer programme registration (figure.ai/developers). Install SDK: pip install figure-sdk. Connect to robot (or simulator): from figure_sdk import Robot; robot = Robot.connect(sim=True). Figure provides a physics simulation (based on MuJoCo) for development without hardware. Basic task: robot.arm.move_to(position=[0.5, 0.0, 1.2], frame="world"). Gripper: robot.hand.grasp(force=30). Access vision: camera_feed = robot.vision.get_frame("head_camera"); detections = robot.vision.detect_objects(camera_feed). Full documentation: developer.figure.ai. Our ML team develops Figure 02 applications.
Figure's Helix model enables natural language task specification: from figure_sdk import Robot, HelixTaskRunner; runner = HelixTaskRunner(robot); result = runner.execute("pick up the red cylinder from the left conveyor and place it in the bin on the right"). Helix interprets the description, queries the vision system to locate objects, plans the grasp and placement, and executes the manipulation sequence. Helix's pre-trained knowledge covers common manufacturing objects and actions; fine-tune on your specific objects and workflows via the Helix fine-tuning API. This capability — describing tasks in natural language rather than programming joint-level motions — is Figure 02's defining developer experience advantage.
Our ML development and software development teams develop applications for Figure 02, Boston Dynamics Atlas, and other humanoid robot platforms. Book a free advisory session.