Helix 02 is the second iteration of Figure AI's proprietary foundation model, presented in June 2025 alongside the Figure 03 humanoid launch. The first version (Helix) was announced in February 2025 and introduced the hierarchical System 1/System 2 architecture: a fast visuomotor controller (~200 Hz) plus a slower language-semantic planner (~7-9 Hz). Helix 02 advances this direction: a 30% smaller model with 40% higher manipulation throughput and 3x longer task sequences.
The Helix 02 architecture has three modules: (1) System 2 — a multimodal VLM (Vision-Language) based on open Llama-3 70B fine-tuned by Figure on humanoid data, decoding goals and plans; (2) System 1 — a 200M-parameter diffusion transformer emitting joint commands within 5 ms; (3) Memory Hub — a short-term buffer of previous observations to maintain context.
Helix 02 training ran on an H200 cluster with 50,000+ hours of recordings from the Figure 02 fleet (~100 units) and synthetic Isaac Sim data. The model is deployed onboard on NVIDIA Jetson Thor and the ARM cluster of Figure 03. Currently used in the BMW Spartanburg production pilot for car assembly.
A Control Stack is the set of software components responsible for control logic, motion planning, command execution, and coordination of a robot's actuators. It receives high-level goals or waypoints (from planning or AI layers) and translates them into low-level actuator commands, closed-loop feedback control, and safety checks. Common implementations include MoveIt (arm manipulation), Nav2 (mobile navigation), and custom whole-body control stacks for humanoids.
A Perception Stack encompasses the software layers that process data from cameras, LiDARs, IMUs, microphones, and other sensors in order to recognise the surrounding environment, perform localisation, detect and track objects, and interpret the scene. It is typically the first processing stage in an autonomous robot's data pipeline, feeding its outputs to planning and control stacks.
A Runtime is the environment or execution layer used to run code, load libraries, manage dependencies, and operate applications or services — either in real time or during normal system operation. In robotics this includes real-time operating system (RTOS) runtimes, ROS 2 executor runtimes, containerised execution environments (Docker, podman), and embedded C++ runtimes on microcontrollers.
Figure 03 at the BMW Spartanburg plant (production pilot since Q1 2025), Figure AI's San Jose facility, prior OpenAI partnership (ended February 2024, continued internally).
No public community — closed model. Figure AI team: 300+ employees, 40+ ML/RL researchers.
Model ~14 GB FP8. System 1 inference 5 ms, System 2 inference 150 ms (planning).
License family: Proprietary – Commercial
Production version used at BMW Spartanburg.
Launched with Figure 03: smaller model, 40% better manipulation.
First generation: System 1/System 2, deployed on Figure 02.
