· Partnerships & M&A· 日本語で読む
RoboSense and BitRobot Announce Strategic Cooperation, Release 100-Hour Work Dataset
RoboSense announced strategic cooperation with BitRobot on Nov. 11 during CoRL 2026. As one result, BitRobot released the 100-hour “RoboCap LiDAR-100,” which records real-world tasks such as kitchen work and auto maintenance. It combines LiDAR, RGB images, inertial measurement data and depth information at QVGA resolution.
What happened
RoboSense announced strategic cooperation with BitRobot on Nov. 11 during CoRL 2026. As one result of the collaboration, BitRobot released “RoboCap LiDAR-100,” which records 100 hours of real-world tasks including kitchen work, auto maintenance and auto detailing.
Why it matters
According to reports on the collection method, LiDAR-derived depth information and inertial measurement data are recorded in sync with RGB images, making the dataset a resource that pairs task footage with spatial measurements. For those researching and developing robot perception and manipulation, the dataset’s terms of use, accuracy assessments and breakdown by task are factors in determining its usability, in addition to the total recording time.
Analysis
The central feature of the announcement is the use of sensors for robot perception in the collection of real-world task data. According to reports, RoboCap LiDAR-100 synchronously records LiDAR, RGB images and inertial measurement data, and outputs depth information at QVGA resolution. The design aims to collect spatial information that includes distance, as well as video, in a single collection process, potentially providing data for hand and head position estimation and environmental reconstruction. However, the article does not confirm which models have actually been trained or evaluated with the data, or whether its effectiveness has been measured.
The 100-hour scale and the focus on kitchen work, auto maintenance and auto detailing indicate the dataset’s scope. However, recording time alone does not show the number of task types, the frequency of each task, the proportion of failure cases or the reproducibility of the data. Reports also do not provide measurement conditions or accuracy assessments showing how much spatial accuracy the combination of multiple sensors offers users. The release of data containing spatial information should therefore be evaluated separately from any improvement in robot manipulation performance.
BitRobot has reportedly previously released datasets with Unitree and Hugging Face. Its cooperation with RoboSense is positioned as an example of incorporating 3D sensing into data collection. This also connects to the industrial-base issues discussed in the Nov. 14 article from leaderobot.com asking which companies support China’s robotics industry, in that corporate cooperation, field data and publicly released data need to be examined separately. In light of the discussion across multiple layers in the Nov. 13 Business+IT article on experts’ views of the “world model” in physical AI, this announcement primarily describes the data-collection setup, not the model approach or training results. The next points to verify are the data’s terms of use, annotation and quality-control methods, breakdown by task, compatible robot configurations, and whether training and evaluation results will be released.
Sources
Translated from the Japanese original: /news/soku-tsukuto-bitrobot-ga-senryaku-teikei-100-jikan-no-3d-sag-9fc241. The Japanese version is authoritative if they differ.