Figure AI showcases upgraded Helix, flips packaging and boxes like a champ.

Helix is another step closer to conquer the factory jobs. Take a look at Helix, the AI powering Figure 02 (the robot below) in handling of packaging sorting on a supply line.

What is Helix?

Helix is Figure AI’s flagship Vision-Language-Action (VLA) model, launched February 20, 2025. Unlike earlier robotics systems that separate perception, planning, and control, Helix unifies all three in a single neural network. It processes camera inputs and natural-language instructions to generate high-level goals, then drives a low-level controller that outputs continuous torque commands for the robot’s entire upper body—including wrists, fingers, torso, and head—completely onboard low-power GPUs. In short, this means that it uses 2 separate systems, that talks to each other, to manage delicate tasks of interacting with soft and hard objects; carefully.

Why this demo is important?

1. Human-level dexterity at speed
   In the latest logistics video (published June 8, 2025), Helix-equipped robots handle a wide variety of package types—flat, boxed, crinkly—without re-training. They’ve cut average handling time from ~5.0 seconds per parcel down to 4.31 seconds, edging closer to (and in some cases matching) human operators.
2. Reliable soft and hard object handling
   Figure, the robot in video, demonstrated reliable ability to handle both soft packaging and hard packaging without damaging it nor incurring unnecessary force.
3. True multi-robot collaboration
   The same set of Helix weights runs simultaneously on two humanoids working side by side—coordinating to lift, reorient, and stow groceries in shared storage—without any task-specific fine-tuning.
4. Zero-shot pick-and-place
   Thousands of novel objects—household knickknacks, industrial parts, you name it—are picked up on first encounter, guided only by simple language prompts like “Grab the blue item”.
5. On-device, real-world readiness
   Everything runs onboard embedded GPUs. No cloud latency, no offsite compute. That makes Helix the first truly commercial-ready VLA for real facilities.

Helping the industry

With labor directly affected by geopolitics, local laws, regulations, population decline and the human-induced problems, it's a no-brainer just how robots can be a much better solution to repeatitive tasks as such.

• Alleviate labor shortages: Logistics and warehousing face chronic staffing gaps—robots like these can fill peak-season surges without expensive temp hiring.
• Boost throughput: Faster, consistent package handling cuts bottlenecks on sorting lines and reduces dwell time in fulfillment centers.
• Lower long-term costs: Once deployed, onboard inference slashes cloud-compute bills and centralizes maintenance under one unified software stack.
• Improve safety: Humanoid form factors navigate human-designed facilities—stairs, narrow aisles, conveyor heights—without bespoke fixtures.

Potential problems still ahead

Figure AI released this recent demo video with a whopping timestamp of an hour, while impressive but throughout the video, various mishaps could be observed. But amazing nonetheless.

1. Speed vs. humans: At 4.31 s per package, Helix is catching up—but on highly optimized lines, human pickers still edge it out. Further cycle-time reduction is critical.
2. Clogging of supply line: In the video you could see a large amount of packages clogging up the line, which were not noticed by the robot. Future improvements and additional robots could help with this issue.
3. Maintenance overhead: Humanoid robots have hundreds of moving joints. Repair and part-replacement costs remain high compared to simpler gantry or articulated-arm systems.
4. Edge-case robustness: Irregular shapes, deformable items (e.g., bags of produce), and tight bin packing still challenge the vision and grasp planner.
5. Regulatory & safety certification: Deploying full-size humanoids around people requires rigorous compliance testing, which can slow rollout.

Which businesses should care?

Any organization wrestling with repetitive pick-and-place or multi-robot coordination stands to gain from adopting a unified VLA like Helix.

• E-commerce & 3PL providers (e.g., Amazon-style fulfillment centers): for dynamic order peaks
• Logistics integrators (FedEx, DHL): to automate parcel sorting and loading
• Retail distribution (grocery, big-box chains): for back-of-house stocking
• Manufacturers with mixed workflows: to handle small-parts kitting in assembly lines
• Hospitality & facilities services: potential down-the-road for room service, housekeeping, and “labour-light” cleaning

We're getting closer to highly reliable humanoid robotics for limited number of factory work, I supposed within the next 3 years, robots like Figure 02 here will be as common as Foldable phones today; you see it here and there but not quite everywhere, yet.