Datasets for research on autonomous robotics
Michigan Robotics offers a number of datasets supporting the development of autonomous systems. One of the most important is YouCook2 - the largest available set of instructional videos, including 2000 uninterrupted recordings from 89 cooking recipes. This dataset is used to test models for action recognition, context understanding, and analysis of activity sequences in real time.
Another key tool is AprilTag - a vision system for fiducial marker identification, used for is used in camera calibration and augmented reality applications. Thanks to its simple interface and stability, it is widely used both in academic research and engineering projects.
Low-cost tools for teaching robotics
Enlarged imageClose zoomPrevious imageAmong the open projects, it is worth mentioning MBot - a low-cost, customizable ecosystem designed for teaching robotics using real devices. The project was created with accessibility in mind for universities and schools, allowing students to gain practical experience in programming and configuring robots.
Similarly, the GelSlim 4.0 project offers a vision-based tactile sensor with a simplified finger structure and easily manufactured lenses. It is designed for research on robot interaction with the environment, especially in the context of manipulating delicate objects.
Resources for robotics researchers and engineers
In addition to datasets, Michigan Robotics provides a range of source code projects. The most important include toolsets for motion planning and manipulation under uncertainty, as well as tools for modeling human leg kinematics. These resources are particularly useful for research teams working on bipedal robots or assistive systems.
Furthermore, data from cameras and lidars from the university campus is available, which can be used to test safety algorithms in autonomous vehicles. All of these resources are available without restrictions, which helps accelerate innovation in the robotics sector.
Datasets for research on autonomous robotics
The YouCook2 dataset, besides its enormous scale, represents an important resource for researchers working on natural language understanding in the context of actions. Thanks to detailed recordings and recipes, models can be trained to interpret sequences of actions, which is crucial for robots operating in home or service environments. This dataset has been used in many studies concerning automatic generation of step-by-step instructions and action goal recognition, contributing to the development of intelligent user assistance systems. Additionally, its continuous recordings allow for testing models under near-real conditions, which increases their reliability and practical application.
AprilTag, as a fiducial identification system, not only supports camera calibration but also finds application in robot localization in 3D space. Thanks to its precision and resistance to lighting changes, it is widely used in laboratory experiments and in real-world projects, such as automatic warehouse systems or industrial robots. Its openness and ease of implementation make it a standard tool in many academic laboratories around the world, which contributes to the unification of research methods and comparability of results.
The datasets and projects from Michigan Robotics not only support the development of technology, but also promote openness and collaboration within the scientific community. Thanks to their unrestricted availability, researchers can compare experimental results, reproduce studies, and build on existing achievements. This contributes to faster progress in the field of autonomous robotics, especially in areas such as action recognition, motion planning, or interaction with the environment.
The YouCook2 dataset, thanks to its unique structure and realistic recordings, is becoming a standard for testing AI models in the context of activity sequences, and AprilTag contributes to stability and precision in 3D localization - key elements for robots operating in dynamic environments. Contemporary projects such as MBot or GelSlim 4.0 show that innovation does not have to be expensive - low-cost solutions can be just as effective in teaching and research as professional systems. In this way, Michigan Robotics creates a foundation for the future of robotics: open, accessible, and focused on practical applications.
