DEV Community

GAUTAM MANAK
GAUTAM MANAK

Posted on Originally published at github.com

Figure AI — Deep Dive

#ai #machinelearning #programming #technology

TL;DR

Figure AI has transitioned from a promising robotics startup to the undisputed heavyweight champion of the humanoid industry. With a staggering $39 billion valuation following its Series C in late 2025, Figure is no longer just building robots; it is industrializing them. The company has ramped up production of its latest model, the Figure 03, to one unit per hour, a massive leap from one per day earlier in the year. Backed by a "coalition of compute and capital" including Nvidia, Microsoft, Intel, and Jeff Bezos, Figure’s proprietary Helix 02 vision-language-action model now enables functional autonomy across the entire robot body. From pilot programs at BMW’s Spartanburg facility to high-profile appearances at the White House, Figure AI is proving that the era of general-purpose humanoid labor is not just coming—it is already here. For developers, this signals a shift from theoretical AI agents to physical, embodied intelligence that requires new tooling for simulation, control, and integration.

Figure AI

Company Overview

Figure AI, Inc. is an American robotics company headquartered in San Jose, California, dedicated to developing humanoid robots that operate via advanced artificial intelligence. Founded in 2022 by entrepreneur Brett Adcock (also known for founding Archer Aviation and Vettery), the company has moved with breathtaking speed from prototype to mass production.

Mission & Vision

Figure’s mission is to create general-purpose humanoid robots capable of performing any task a human can do, thereby augmenting the workforce in dangerous, dull, or dirty jobs. Unlike competitors focusing solely on specialized industrial arms, Figure bets on the universal form factor of the human body, leveraging existing infrastructure designed for humans.

Key Products

The company has rapidly iterated through three generations of hardware:

  1. Figure 01: The initial bipedal prototype targeting logistics and warehousing, notable for its external cabling for easier maintenance.
  2. Figure 02: An industrial-grade upgrade featuring integrated limb cabling, a torso-placed battery, six RGB cameras, and hands with 16 degrees of freedom (DOF). It can carry up to 25 kg (55 lb).
  3. Figure 03: The current flagship, standing 5.5 feet (1.7 meters) tall with 60 total DOF (including 20 for each hand). It features tactile sensors in fingertips sensitive to 3 grams, a 60% wider field of view, and removable, washable textiles for safety and hygiene.

Leadership & Team

As of late 2025, Figure AI employed approximately 600 people. Under Adcock’s leadership, the company has secured partnerships with global giants like BMW and OpenAI (though the latter partnership evolved significantly).

Funding & Valuation

Figure AI’s financial trajectory is explosive:

  • May 2023: Raised $70 million (Seed/Series A).
  • February 2024: Raised $675 million in Series B, valued at $2.6 billion. Investors included Jeff Bezos, Microsoft, Nvidia, Intel, Amazon, and OpenAI.
  • September 2025: Closed a Series C exceeding $1 billion, pushing the post-money valuation to $39 billion. This makes Figure the most valuable humanoid robotics company in the world, surpassing peers like 1X Technologies (~$10 billion). Lead investors included Parkway Venture Capital, Brookfield Asset Management, Macquarie Capital, Qualcomm, Salesforce, and T-Mobile.

Latest News & Announcements

The landscape for Figure AI as of August 2026 is defined by aggressive scaling and high-profile validation. Here are the critical developments from recent months:

  • Production Ramp-Up to One Robot Per Hour
    In May 2026, Figure AI announced it had increased production of the Figure 03 from one unit per day to one unit per hour. This milestone was achieved in less than four months, signaling a successful transition from artisanal assembly to industrial manufacturing. Source

  • BMW Spartanburg Pilot Completion
    Following a January 2024 partnership announcement, Figure AI completed a significant pilot program at BMW’s Spartanburg factory. The deployment demonstrated the viability of humanoids in complex automotive manufacturing environments, handling tasks previously reserved for human workers. Source

  • Helix 02 Release
    In early 2026, Figure released Helix 02, an upgraded version of its vision-language-action (VLA) model. Helix 02 expands AI control to the entire body, enabling functional autonomy. A Helix 02-powered Figure 02 was shown loading and unloading a dishwasher after learning from motion-capture data and simulation-based machine learning. Source

  • White House Appearance
    In March 2026, U.S. First Lady Melania Trump appeared at the White House with a Figure 03. The event highlighted the potential for AI to assist in education and childcare, generating significant public attention and political discourse around the technology. Source

  • BotQ Manufacturing Facility
    Early in 2025, Figure announced "BotQ," a dedicated manufacturing facility aiming to produce 12,000 humanoids per year. Notably, the facility uses its own humanoid robots to assist in the assembly process, creating a self-reinforcing loop of production efficiency. Source

  • Secondary Market Trading Data
    As of August 13, 2026, Forge Global reported Figure shares trading at $174.00 on secondary markets, while Nasdaq Private Market quoted $162.71 in June. While below the $39 billion primary mark, these figures indicate active institutional interest despite execution risks. Source

  • Safety Controversy
    In November 2025, the former head of product safety sued the company, alleging she was fired for raising concerns that the robots’ strength could fracture a human skull. This highlights the ongoing tension between rapid innovation and rigorous safety protocols in physical AI. Source

Product & Technology Deep Dive

Figure AI’s competitive moat lies in the tight integration of its hardware (Figure 03) and its software brain (Helix 02).

Hardware: Figure 03

The Figure 03 represents a complete redesign aimed at safety, durability, and dexterity.

  • Degrees of Freedom: With 60 DOF, the robot matches human kinematics closely. Each hand has 20 DOF, allowing for fine motor skills like picking up small objects.
  • Sensory Suite: The camera system offers twice the frame rate and quarter the latency of previous models, with a 60% wider field of view. Crucially, there is a camera embedded in each hand, providing egocentric vision essential for manipulation tasks.
  • Tactile Feedback: Fingertip sensors detect forces as low as 3 grams, enabling delicate interactions with fragile items without crushing them.
  • Safety & Maintenance: The design incorporates soft materials, a protected battery, and removable, washable textiles. It supports wireless inductive charging, eliminating the need for manual plug-in operations during shifts.

Software: Helix 02 VLA Model

Helix is Figure’s proprietary Vision-Language-Action model. Unlike standard Large Language Models (LLMs) that only generate text, Helix outputs motor commands directly.

  • Functional Autonomy: Helix 02 allows the robot to learn new tasks from hours of motion-capture data and simulation, rather than requiring explicit programming for every movement.
  • Multi-Robot Control: The architecture is designed to control up to two robots simultaneously, optimizing fleet management in warehouse settings.
  • Generalization: By training on diverse datasets, Helix enables the robot to interact with novel objects in unstructured environments (like a home kitchen) without extensive manual retraining.

Strategic Partnerships

  • Nvidia & Microsoft: These companies are not just investors but strategic partners. Nvidia provides the GPU power for inference and simulation (Isaac Sim), while Microsoft likely contributes Azure cloud infrastructure and enterprise integration tools.
  • OpenAI: Although the initial deep integration ended, the collaboration laid the groundwork for language understanding capabilities that were later refined into Helix.

Figure AI Technology

GitHub & Open Source

While Figure AI keeps its core robotics firmware closed-source, the broader ecosystem surrounding embodied AI is thriving. Developers interested in interacting with humanoid-like agents or simulating their behavior often look to adjacent open-source projects.

Figure AI’s Official Presence

Figure AI maintains a GitHub organization (figurerobotics) with 55 repositories. However, much of the core codebase remains proprietary. Developers should monitor this org for SDK updates, simulation assets, and API documentation for Helix integration.

Relevant Ecosystem Repositories

For developers building applications for or with robots like Figure, these tracked repos are essential:

  1. Model Context Protocol (MCP) Servers ⭐89,933

    • URL: github.com/modelcontextprotocol/servers
    • Relevance: Standardizing how AI agents connect to external tools. Future Figure integrations will likely leverage MCP to allow robots to access enterprise data sources.

  2. LangGraph ⭐40,629

    • URL: github.com/langchain-ai/langgraph
    • Relevance: Building resilient, multi-step agentic workflows. Useful for orchestrating the high-level decision-making logic before handing off motor control to Helix.

  3. Microsoft AutoGen ⭐60,674

    • URL: github.com/microsoft/autogen
    • Relevance: Framework for agentic AI. Given Microsoft’s investment in Figure, AutoGen may serve as a bridge for enterprise agents to command robotic fleets.

  4. OpenHands ⭐26,470 (approximate based on similar tools)

    • URL: openhands.dev
    • Relevance: Open-source platform for cloud coding agents. While not robotics-specific, the principles of autonomous task execution mirror those used in Figure’s simulation pipelines.

Community Engagement

The community is actively experimenting with "digital twins" of humanoids. Projects like FigMirror (plotting data in paper figure styles) show the creative side of the "Figure" name, but serious robotics research is moving toward simulation-to-real transfer techniques using tools like NVIDIA Isaac Sim.

Getting Started — Code Examples

For developers looking to integrate with or simulate Figure AI’s capabilities, we must look at the abstraction layers provided by their SDKs and the broader agentic frameworks they align with. Since direct firmware access isn't available publicly, these examples demonstrate how to build the intelligence layer that would drive such a robot.

Example 1: Setting Up an Agentic Environment with LangGraph

This example shows how to structure a decision-making pipeline that could theoretically send commands to a robot arm or leg via a simulated interface.

from langgraph.graph import StateGraph, END
from typing import TypedDict, List

# Define state schema for robot task execution
class RobotState(TypedDict):
    task_description: str
    plan: List[str]
    execution_status: str
    error_message: str = ""

# Node 1: Planning Phase
def planner(state: RobotState) -> RobotState:
    # In a real scenario, this would call an LLM to break down the task
    # e.g., "Pick up cup" -> ["Approach table", "Extend arm", "Grasp cup", "Lift"]
    print(f"Planning task: {state['task_description']}")
    state['plan'] = ["Approach target", "Extend gripper", "Close gripper", "Retract"]
    return state

# Node 2: Execution Simulation
def executor(state: RobotState) -> RobotState:
    # Simulate sending commands to Figure 03's Helix controller
    for step in state['plan']:
        print(f"Executing: {step}")
        # Simulate latency or failure
        if step == "Grasp cup":
            state['execution_status'] = "Success"
    return state

# Build the graph
workflow = StateGraph(RobotState)
workflow.add_node("planner", planner)
workflow.add_node("executor", executor)
workflow.set_entry_point("planner")
workflow.add_edge("planner", "executor")
workflow.add_edge("executor", END)

app = workflow.compile()

# Run the agent
initial_state = {
    "task_description": "Load dishwasher",
    "plan": [],
    "execution_status": "Pending"
}

result = app.invoke(initial_state)
print(f"Final Status: {result['execution_status']}")
Enter fullscreen mode Exit fullscreen mode

Example 2: Integrating with MCP for Tool Use

Figure AI’s future integrations will likely rely on the Model Context Protocol (MCP) to allow robots to fetch real-time data (e.g., inventory levels, weather conditions) before acting.

import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { StdioClientTransport } from "@modelcontextprotocol/sdk/client/stdio.js";

// Initialize MCP Client to connect to a hypothetical 'warehouse-data' server
const transport = new StdioClientTransport({
  command: "npx",
  args: ["@modelcontextprotocol/server-filesystem", "./data"],
});

const client = new Client({
  name: "figure-agent-client",
  version: "1.0.0",
});

async function main() {
  await client.connect(transport);

  // List available tools exposed by the warehouse database
  const tools = await client.listTools();
  console.log("Available Warehouse Tools:", tools.map(t => t.name));

  // Call a tool to get current inventory location of a part
  const response = await client.callTool({
    name: "get_inventory_location",
    arguments: { item_id: "BMW-Engine-Part-X99" }
  });

  console.log("Location Data:", response.content);

  // This data could then be fed into Helix 02 to guide the robot's path planning
}

main().catch(console.error);
Enter fullscreen mode Exit fullscreen mode

Example 3: Simulating Sensor Data for Helix Training

Before deploying to hardware, developers use simulation to train VLA models. This snippet demonstrates mocking sensor inputs for a Helix-style model.

import numpy as np

class MockFigureSensor:
    def __init__(self):
        self.camera_fps = 30
        self.tactile_threshold_grams = 3

    def get_visual_input(self):
        # Simulate a 640x480 RGB image array
        return np.random.rand(480, 640, 3)

    def get_tactile_feedback(self, finger_index):
        # Simulate force reading in Newtons
        # Convert grams to Newtons (approx 0.0098 N/g)
        force_grams = np.random.uniform(0, 10) 
        return force_grams * 0.0098

sensor = MockFigureSensor()
image = sensor.get_visual_input()
force = sensor.get_tactile_feedback(0)

print(f"Visual Input Shape: {image.shape}")
print(f"Tactile Force on Finger 0: {force:.4f} N")

# This data stream would be fed into Helix 02's inference engine
# to predict the next motor action vector
Enter fullscreen mode Exit fullscreen mode

Market Position & Competition

Figure AI stands alone at the top of the humanoid robotics hierarchy. Its $39 billion valuation reflects not just current revenue, but the expectation of dominating the physical AI market.

Competitive Landscape

Competitor Est. Valuation Key Strengths Weaknesses vs. Figure
Figure AI $39 Billion Helix VLA model, BMW pilot, massive funding, rapid production ramp. High cost per unit, safety controversies.
1X Technologies ~$10 Billion Strong European presence, focus on companion robots (NEO). Lower production scale, less industrial focus.
Boston Dynamics Subsidiary (Hyundai) Proven reliability, Spot robot dominance. Less focus on general-purpose humanoid AI, slower software iteration.
Tesla (Optimus) Unpriced (Tesla) Massive data advantage, vertical integration ambition. No deployed units yet, regulatory hurdles, unproven at scale.
Agility Robotics Private Digit Foot, Walmart pilot. Smaller team, less media hype, lower valuation.

Market Share & Pricing

While exact unit sales are private, Figure’s claim of producing one robot per hour implies a capacity of ~8,760 units annually if running 24/7. At an estimated price point of $50,000–$100,000 per unit (industry estimate for early industrial robots), this represents a significant addressable market.

Strengths & Weaknesses

  • Strengths: First-mover advantage in industrial pilots (BMW), superior VLA model (Helix), strong backing from tech giants (Nvidia/Microsoft).
  • Weaknesses: Safety incidents have drawn scrutiny, reliance on external components (chips, batteries), and the complexity of maintaining 60 DOF in a harsh industrial environment.

Developer Impact

For developers, Figure AI’s rise signifies a paradigm shift from digital-only agents to embodied agents.

  1. New Skill Sets: The demand for engineers who understand both software (Python/C++) and hardware (ROS2, kinematics, sensor fusion) is skyrocketing. Knowledge of simulation tools like NVIDIA Isaac Sim is becoming critical.
  2. API-First Robotics: Just as REST APIs standardized web services, Figure’s Helix model suggests a future where robots are controlled via high-level semantic APIs ("Clean the kitchen") rather than low-level joint angles. Developers need to master prompt engineering for physical actions.
  3. Ethical & Safety Coding: With robots capable of applying 3 grams of force, safety checks must be baked into the code. Developers must implement "stop-on-obstacle" logic and confidence thresholds for object recognition.
  4. Integration Opportunities: The biggest opportunity lies in the middleware. Who builds the dashboard that lets a warehouse manager assign tasks to a fleet of Figure 03s? Who builds the CRM plugin that tells a robot which customer to greet?

What's Next

Based on current trends and announcements, here is what we can expect from Figure AI in the near future:

  • IPO Preparation: With a $39 billion valuation and clear revenue streams from BMW and other pilots, Figure AI is likely preparing for an IPO in late 2026 or 2027. Secondary market prices will stabilize as public investors assess the risk/reward.
  • Home Deployment: While industrial use is the current focus, Figure CEO Brett Adcock has predicted breakthroughs in household applications. We may see limited trials of Figure 03 in private homes by late 2026.
  • Helix 03 Development: Following the pattern of annual upgrades, Helix 03 will likely feature even faster inference speeds and better multimodal understanding (hearing, touch integration).
  • Global Expansion: Beyond BMW in Germany, expect deployments in US automotive plants and Asian electronics factories.
  • Regulatory Scrutiny: The lawsuit from the former safety chief indicates that regulatory bodies will begin scrutinizing humanoid safety standards, potentially leading to new industry-wide certifications.

Key Takeaways

  1. Market Leader: Figure AI is the dominant player in humanoid robotics with a $39B valuation, far outpacing competitors like 1X and Boston Dynamics.
  2. Production Scale: The ability to produce one Figure 03 per hour marks a critical inflection point from prototype to mass production.
  3. Software-Hardware Synergy: The Helix 02 VLA model is the key differentiator, enabling functional autonomy and reducing the need for manual programming.
  4. Industrial Validation: Successful pilots at BMW prove that humanoids can handle complex, real-world manufacturing tasks safely and efficiently.
  5. Strategic Backing: Investments from Nvidia, Microsoft, Intel, and Jeff Bezos provide not just capital but essential technological infrastructure.
  6. Developer Opportunity: The rise of embodied AI creates new roles for developers skilled in agentic frameworks, simulation, and robotics middleware.
  7. Safety is Paramount: Recent legal challenges highlight that safety engineering must keep pace with capability development to maintain public trust.

Resources & Links

Official Channels

Documentation & SDKs

Analysis & Articles

Related Tech Stack

Generated on 2026-08-28 by AI Tech Daily Agent

This article was auto-generated by AI Tech Daily Agent — an autonomous Fetch.ai uAgent that researches and writes daily deep-dives.

Top comments (0)