US-based robotics company Figure AI announced that its Figure 03 humanoid robots, powered by the new generation Helix-02 artificial intelligence system, can now work fully autonomously for durations similar to humans. According to the company's shared information, the robots can complete full 8-hour work shifts without any human intervention. When a shift ends, it is handed over to another robot.
Broadcast Live
Figure AI stated in a post on X that a group of humanoid robots performed tasks for eight hours at "human performance levels." The company specifically emphasized that the system operates completely autonomously. The company broadcast the entire work shift live, and the broadcast is ongoing.
As of the time this content was written, the robots had been working for over 14 hours and had processed more than 18,000 packages. A 24-hour broadcast is expected. The working team consists of robots named Bob, Frank, and Gary.
In the test setup, the robots detect packages arriving on a conveyor belt in an environment resembling a logistics center and reposition them with their barcodes facing downwards. According to Figure AI's statement, the entire process is managed by the company's developed Helix-02 neural network system. It is emphasized that no human control is needed throughout the operation.
21 Packages Processed Per Minute
Currently, the robots process 21 packages per minute. This means an average processing time of approximately 2.8 seconds per package. Company CEO Brett Adcock had previously announced that the robots achieved "human-level performance" and that the processing speed dropped below three seconds. The live broadcast aims to demonstrate this claim in real-time.
The packages processed by the robots were not limited to standard boxes. The footage showed cardboard boxes of various sizes, as well as flexible colored plastic packages. Figure AI states that this diversity demonstrates the Helix-02 system's ability to detect and process different objects.
End-to-End Artificial Intelligence is Used
One of the most important details highlighted by the company was the software architecture used. Figure AI employs an end-to-end learning approach, which it calls "Software 2.0," instead of the hand-written C++ based motion planning codes common in classic robotic systems.
[An embedded tweet from Brett Adcock was shown here.]
In this system, the AI model directly calculates the torque values to be applied to the motors, managing the robot's movements. Thus, the robot can perform object detection, grasping, orientation, and motion coordination under a single learning system.
Errors Were Also Observed During the Broadcast
[An embedded tweet from RoseOnX9 was shown here.]
The system did not operate flawlessly throughout the live broadcast. In the early stages, one of the robots was seen incorrectly orienting a cardboard box, with the barcode not facing the desired direction. There were also brief pauses at times, and some packages were missed. However, according to Figure AI, the crucial point was that the system did not completely stop after an error and continued its task.
The company argues that occasional minor errors are natural in long-term autonomous operations, and what matters is that the robots can continue their tasks without system crashes.
Robots Can Self-Diagnose Faults and Request Replacement
[An embedded tweet from Brett Adcock was shown here.]
Brett Adcock had acknowledged in statements before the live broadcast that there was a high probability of technical issues during the eight-hour continuous operation. For this reason, the company developed a special structure called "multi-robot coordination."
According to Adcock's explanation, when a robot detects a systemic problem, it can self-diagnose. If the problem persists, the robot autonomously proceeds to the maintenance area and requests another robot in the fleet to take over the task. No human intervention is involved in this entire process.
Although the broadcast was technically impressive, some experts considered the scope of the demonstration limited. The core of the criticism was the robots' focus on only a single task. Performing the same package sorting operation for eight hours was not seen as sufficiently comprehensive by some commentators in terms of a general-purpose humanoid robot vision. However, Figure AI's objective is considered to be demonstrating the robots' physical endurance and their ability to work long shifts without human support.
Source: https://interestingengineering.com/ai-robotics/figure-helix02-humanoid-robots-8-hour-shifts
0 Comments: