TMTPost publishes article on technical approaches to embodied intelligence

TMTPost published an article on technical approaches to embodied intelligence on May 20, 2026.

The publication of a media article alone does not confirm that one training method is replacing another or that real-world robot performance has improved. Evaluators need to examine the scale of the data infrastructure, performance on real-world tasks and the specifications of the sensors used separately, and determine whether the conditions being compared are consistent.

The material concerns the publication of a media article about debate over robot learning methods. Even if names for technical approaches gain broad use, they do not by themselves allow comparison of robots' capabilities or areas of application. Evaluation requires separately examining the environment, the operations performed and the data used for training and validation. For details of the article's argument and statements, consult the original. Related articles in this newspaper illustrate technical challenges at different levels. [Hugging Face releases robot-learning platform LeRobot v0.7.0, supporting conversational operations and large-scale data training](/en/news/hugging-face-releases-lerobot-v0-7-0-robotics-training-platform-adding-4e8d3f) covered a platform that handles about 300TB of data and supports functions from training through execution. Meanwhile, [Panasonic Holdings and others jointly develop robot manipulation AI T-Rex, with an average success rate of 65% across 12 tasks](/en/news/panasonic-holdings-partners-develop-t-rex-robot-manipulation-ai-65-ave-275dc6) reported results on real-world tasks using tactile sensing and described the composition of the training data. The two are not the same kind of evaluation, but they show why the scale of a data platform should not be conflated with performance on individual manipulation tasks. In addition, the camera module covered in Asumekku begins handling 8.3MP GMSL2 camera modules is a component in the system robots use to observe their surroundings. Sensors, manipulation models and data-collection and training platforms serve different stages; a model's name alone cannot explain the overall performance. These reports also do not directly prove the superiority of any technical approach. For the industry, the focus is not only on the amount of training data, but also on which sensors and operating conditions produced it and how closely they match the environment used for evaluation. Points to check going forward include comparisons under the same tasks and conditions, recovery from failures, sustained operation in real-world environments, and the composition and terms of use of training data. The public information available does not necessarily allow these factors to be compared on a like-for-like basis.