University of Tokyo Research Team Equips Musashi-W With Biomimetic Skin to Estimate Joint Angles

According to a Sept. 13, 2026, report by leaderobot.com, a University of Tokyo research team fitted Musashi-W, a musculoskeletal humanoid robot, with three-layer biomimetic joint-covering skin and tested a method for estimating joint angles using only information from skin deformation. The report said the mean estimation error was about 3 degrees, and accuracy improved further when skin and muscle information were combined.

Because of their structure, musculoskeletal robots cannot directly measure joint angles as easily as robots with mechanical joints. The University of Tokyo report is significant because it suggests that a design sensing skin deformation could also be used to estimate joint states despite this constraint.

The novelty of this work is that it treated the skin itself as an input for proprioception, rather than basing joint-angle estimates on muscle length or tension. However, based on the report alone, this is research that expands how a robot can obtain information about its body; it does not indicate a mass-production capability or commercialization at this stage. Consult the original source for details on the configuration and experimental conditions. Because this newspaper has no related articles, we cannot directly connect this work with previous coverage here. But in the context of publicly available information, it has long been known that musculoskeletal robots face constraints on direct angle measurement compared with robots with mechanical joints. This report appears significant in showing a design that gathers information across distinct layers, such as skin, deeper tissues and muscles, rather than relying on a single high-performance sensor. In particular, if signals from the skin can be used to read angles with an error of about 3 degrees, designers of similar robots may have reason to reconsider the assumption that muscle sensors alone are sufficient. At the same time, the information available does not show how durable the system is in operation, how errors vary under different movement conditions, or how broadly the fusion of skin and muscle information can be applied. A demonstration on a research robot such as Musashi-W does not mean it can be transferred directly to robots with different structures. The focus therefore shifts to three questions: (1) how reproducible the skin material, thickness and receptor placement are; (2) how generalizable the training conditions for fused estimation are; and (3) whether deformation characteristics remain stable during long-term operation. For experimental conditions and results relevant to the article’s subject, it is appropriate to consult the original source.

In Japan’s robotics research, the design of real-world data that integrates tactile, skin and muscle information is likely to be important. The findings are relevant to companies and research institutions with expertise in skin materials, flexible sensors and motion measurement, but publicly available information does not confirm any specific commercial deployment.