Physical AI News (Japan)

Physical AI, robotics and compute — reported from Japan.

日本語
· Tenders· 日本語で読む

APTO selected for student-focused physical AI challenge in GENIAC-PRIZE 2026

APTO will run an open-entry challenge in GENIAC-PRIZE 2026. Students will work on a robot-arm task by devising ways to design data for imitation learning. The preliminary round will be held online, and the final will use real machines at Toyosu Data Factory.

Key facts

  • APTO announced that it had been selected for an open-entry challenge in Theme 2 of GENIAC-PRIZE 2026.
  • The preliminary round will be held in an online simulation environment, and the final will use OpenArm/SO-101 at Toyosu Data Factory.
  • The challenge is seeking up to 20 participants across five teams, with up to three teams advancing to the final. Teams may have up to four members.
  • APTO says it will provide participants with a development environment and computing resources.
  • The prize for passing the preliminary round is ¥30,000 per person. In the final, first place will receive ¥100,000 per person and second place ¥60,000 per person.

What happened

APTO announced Nov. 18, 2026, that it had been selected for Theme 2 of GENIAC-PRIZE 2026, run by the Ministry of Economy, Trade and Industry and NEDO, with its open-entry challenge, “Physical AI Challenge: Compete Through Data Design.” Students may participate individually or in teams of up to four. The challenge is seeking up to 20 participants across five teams. The application deadline is noon Nov. 21, 2026, and up to three teams will advance to the final.

Why it matters

APTO says it will provide participants with a development environment and computing resources, and will run the challenge from an online preliminary round through a final with real machines. Participation is limited to students who pass screening, but the format allows participants without their own GPUs to take part in the preliminary round. It offers an opportunity to connect data design with the evaluation of a real-world task.

Analysis

The challenge focuses not on competing over robot model architectures, but on how data for imitation learning is designed, collected and selected, and how it is used for training and evaluation. APTO will provide a computing environment for the preliminary round, while the final will evaluate real machines, OpenArm/SO-101, at Toyosu Data Factory. A distinctive feature of the format is that it incorporates both simulated work and verification with real machines in a single challenge. The challenge is limited to up to 20 participants across five teams, with up to three teams advancing to the final. It is open only to students who pass screening for Theme 2 and have development experience with Python/PyTorch. It is therefore not a format open to the general public, but a practical opportunity for a limited pool of participants to connect differences in data design to task results. APTO says it will also provide computing resources, so owning a GPU is not a participation requirement. Meanwhile, the stated availability of up to ¥400 million worth of computing resources for Theme 2 as a whole is separate from information about the scale and allocation of resources APTO will provide for this challenge, and the two should not be conflated. A related article from this newspaper covers the distribution of equipment used for data collection: HELTEC named domestic sales distributor for Melt Interface Technologies data gloves. However, the available information does not establish a connection between the data gloves in that article and this arm challenge. The information states only that APTO’s facility, where it works on collecting manipulation data through remote operation, will serve as the venue for the final. It does not establish that any particular input device or data format will be used. The industry significance is that the challenge addresses the practical work of data design, which precedes model performance evaluation, in both simulation and real-world settings. The competition can test whether performance in simulation holds up with real machines when data-generation methods change, but it does not show whether the results would apply to other robots or tasks. The published materials do not specify the task, data volume or format, scoring method for accuracy, evaluation conditions for the final, or how the resulting data and outcomes will be handled. Whether these points will be explained in a comparable form after the challenge is completed remains to be seen.

Details

Participants will devise ways to design, collect and select data for imitation learning, then evaluate learning results on a robot-arm task. The preliminary round will take place in an online simulation environment, with APTO providing the development environment and computing resources. The final will use real OpenArm/SO-101 machines at APTO’s Toyosu Data Factory. The preliminary round is scheduled for Dec. 14, 2026, through Jan. 6, 2027, and the final for Jan. 30-31, 2027. The prize for passing the preliminary round is ¥30,000 per person. In the final, each member of the first-place team will receive ¥100,000, and each member of the second-place team will receive ¥60,000. Participants must be students who pass Theme 2 screening and have development experience with Python/PyTorch.

Sources

Translated from the Japanese original: /news/apto-geniac-prize-2026-no-gakuseimuki-ke-fijikaru-ai-kadai-n-eed3b4. The Japanese version is authoritative if they differ.