Figure Helix 02: The Robot That Loads Dishwashers Without Human Help

In early February 2026, Figure AI released a demonstration video that sent ripples through the robotics industry. Their Helix 02 robot AI software, running on the Figure 03 humanoid, completed a 4-minute dishwasher loading and unloading task—fully autonomously, in a real kitchen, without any human intervention.

The Demo That Changed Everything

Unlike carefully staged product launches, Figure's demo was remarkably mundane—and that's exactly what made it extraordinary. The robot:

1. Walked to the dishwasher
2. Opened the door using its foot
3. Unloaded clean dishes to cabinets
4. Loaded dirty dishes from the sink
5. Closed a drawer with its hip (hands full)
6. Walked away

No teleoperator. No edits. No scripted sequences. Just a robot doing chores.

┌────────────────────────────────────────────────────────────────────┐
│                     HELIX 02: SYSTEM OVERVIEW                       │
├────────────────────────────────────────────────────────────────────┤
│                                                                    │
│   ┌──────────────────────────────────────────────────────────┐    │
│   │                    SENSOR FUSION                          │    │
│   ├────────────────┬─────────────────┬───────────────────────┤    │
│   │   Vision       │    Tactile      │   Proprioception      │    │
│   │   (Cameras)    │   (Touch)       │   (Joint Position)    │    │
│   └───────┬────────┴────────┬────────┴──────────┬────────────┘    │
│           │                 │                   │                  │
│           ▼                 ▼                   ▼                  │
│   ┌────────────────────────────────────────────────────────────┐  │
│   │           UNIFIED VISUOMOTOR NEURAL NETWORK                │  │
│   │                                                            │  │
│   │   Pixels ────────────────────────────────────▶ Actions    │  │
│   │                                                            │  │
│   │   • End-to-end learning                                    │  │
│   │   • No symbolic planning                                   │  │
│   │   • Direct sensor-to-actuator mapping                      │  │
│   └────────────────────────────────────────────────────────────┘  │
│           │                                                        │
│           ▼                                                        │
│   ┌────────────────────────────────────────────────────────────┐  │
│   │                    WHOLE BODY CONTROL                      │  │
│   │                                                            │  │
│   │   ┌─────┐  ┌─────┐  ┌─────┐  ┌─────┐  ┌─────┐  ┌─────┐   │  │
│   │   │Head │  │Arms │  │Hands│  │Torso│  │Legs │  │Feet │   │  │
│   │   └─────┘  └─────┘  └─────┘  └─────┘  └─────┘  └─────┘   │  │
│   │               All actuators controlled simultaneously      │  │
│   └────────────────────────────────────────────────────────────┘  │
│                                                                    │
└────────────────────────────────────────────────────────────────────┘

What Is "Pixels to Whole Body" Control?

The breakthrough technology behind Helix 02 is its unified visuomotor neural network—a single AI system that:

• Takes raw camera pixels as input
• Outputs motor commands to every joint
• Operates end-to-end without symbolic reasoning

Traditional Robotics vs. Helix 02

Why This Matters

Previous robots required engineers to:

1. Detect objects (plates, cups, utensils)
2. Build 3D models of the scene
3. Plan a sequence of actions
4. Execute each action with a separate controller
5. Handle errors with explicit fallback logic

Helix 02 learns the entire task end-to-end. No explicit object recognition, no symbolic planning, no separate controllers.

Technical Achievements

1. Long-Horizon Loco-Manipulation

The dishwasher task requires continuous multi-minute operation combining:

• Walking while carrying delicate objects
• Maintaining stable grasps during locomotion
• Navigating around furniture and obstacles
• Transitioning between manipulation and movement

This is the longest autonomous manipulation sequence demonstrated by a humanoid robot.

2. Human-Like Emergent Behaviors

The robot learned behaviors that weren't explicitly programmed:

• Foot-door opening: Using its foot to open the dishwasher when hands were occupied
• Hip-drawer closing: Bumping a drawer closed with its hip
• Compensatory balance: Adjusting posture when carrying uneven loads

These emerged from training on 1,000+ hours of human motion data combined with sim-to-real reinforcement learning.

3. Integrated Tactile Sensing

Figure 03's hardware includes:

• Palm cameras: Visual feedback for grasping
• Tactile sensors: Pressure detection across fingers
• Force feedback: Real-time grip adjustment

# Conceptual representation of sensor integration
class HelixController:
    def grasp(self, target):
        while not self.tactile.stable_grip():
            visual_target = self.palm_camera.locate(target)
            grip_force = self.predict_grip(visual_target)
            self.fingers.adjust(grip_force)
            
            # Whole-body compensation
            if self.balance.unstable():
                self.legs.compensate()

How Helix 02 Was Trained

Simulation-to-Reality Transfer

1. Human Motion Capture: 1,000+ hours of humans performing household tasks
2. Physics Simulation: Digital twin environments for safe exploration
3. Reinforcement Learning: Billions of simulated episodes
4. Domain Randomization: Varied lighting, textures, object shapes
5. Real-World Fine-Tuning: Final calibration on physical hardware

The Data Pipeline

┌─────────────────┐     ┌─────────────────┐     ┌─────────────────┐
│  Human Motion   │────▶│   Simulation    │────▶│   Real Robot    │
│    Capture      │     │   Training      │     │   Deployment    │
└─────────────────┘     └─────────────────┘     └─────────────────┘
       │                       │                       │
       ▼                       ▼                       ▼
  1000+ hours             10B+ episodes           Continuous
  of demonstrations       explored                learning

Comparison: Figure vs. Tesla Optimus vs. Boston Dynamics

What This Means for the Future

Near-Term (2026-2028)

• Warehouse logistics: Robots picking and packing orders
• Manufacturing: Assembly line assistance
• Elderly care: Home assistance robots (supervised)

Medium-Term (2028-2030)

• Home robots: General-purpose household assistants
• Hospitality: Hotel and restaurant service
• Healthcare: Patient transport and basic care

Long-Term (2030+)

• Universal robots: Adaptable to any physical task
• Personal assistants: Affordable home robots
• Labor transformation: Significant workforce implications

The Business Case

Figure AI has raised $750 million at a $2.6 billion valuation as of late 2025. Their business model:

1. B2B First: Enterprise customers in logistics and manufacturing
2. Learn from Deployment: Data from commercial use improves models
3. Consumer Later: Home robots once technology and pricing mature

Revenue Projections (Figure's Investor Deck)

Skeptical Takes

Not everyone is convinced:

"The dishwasher demo is impressive, but it's one task in one kitchen. Real homes have infinite variations. The generalization problem remains unsolved." — Robotics researcher, MIT

"We've seen amazing demos before. The question is reliability—can this run 8 hours a day, 7 days a week, without supervision?" — Automation industry analyst

Open Questions

1. Generalization: Can Helix 02 handle kitchens it wasn't trained on?
2. Edge Cases: What happens when a glass slips or a door jams?
3. Maintenance: How often does hardware fail?
4. Safety: What protections exist against harm to humans?

Figure's Roadmap

Final Thoughts

Figure's Helix 02 demonstration represents the most convincing evidence yet that humanoid robots are approaching practical utility. The shift from scripted demos to genuine autonomous behavior—however limited—is a qualitative leap.

Whether this translates to robots in our homes within five years remains uncertain. But for the first time, it feels plausible.

Rating: 4.5/5 ⭐

Groundbreaking technology with a clear path to commercialization. The humanoid robot future is closer than most realize.

Related Reading

• Multi-Agent AI Systems Explained 2026
• The Rise of Agentic AI 2026

What household tasks would you want a robot to handle first? Share your thoughts in the comments.