making science happen
Musculoskeletal robot platform
a full scale humanoid robot research platform
Anthropomimetic robots sense, behave, interact, and feel like humans. By this definition, they require human-like physical hardware and actuation but also brain-like control and sensing. The most self-evident realization to meet those requirements would be a human-like musculoskeletal robot with a brain-like neural controller. In traditional approaches, the controller, by its internal functionality, finds appropriate actions on the basis of the history of sensor values, guided by the goals, intentions, objectives, learning schemes, and so forth.
Musc is the only open-source modular musculoskeletal robotics toolkit. Musc shows comparable behavior to biological musculoskeletal systems. It has all the complex effects – pull only, tendon wrapping, hysteresis – that biological muscles have as well. Use it to simulate the effects of a change in muscle attachment or muscle strength. They are therefore ideally suited for a bio-plausible control pipeline in locomotor research. The bioinspired mechatronics are ideally suited as a basis to build prosthetics and exoskeletons based on this technology.
Current vision architecture focuses mainly on object recognition and object tracking. It is using Intel Movidius Neural Compute Sticks, which enable high performance at low cost, specifically does not require a GPU. The Audio processing module is designed to function in the noisy environment. In order to achieve this, the circular microphone array is used, which combined with the beamforming algorithms from ODAS and Gaussian Mixture Models guarantees stable speech recognition, speaker localization and diarization.
Why musculoskeletal Robotics?
scalability in neural control
Compliant, musculoskeletal robotic systems offer several advantages, especially in situations where human and robot work in close proximity. A musculoskeletal design makes extensive use of viscous-elastic materials to emulate the muscles and tendons which enhance safety, dexterity and adaptivity in uncertain environments. It also allows reducing body weight and developmental cost, while at the same time increasing design flexibility.
Robots that mimic the mechanical properties of the
human build strive toward both attributes simultaneously, as, by design, they possess built-in compliance and relatively natural. Musculoskeletal robots in particular offer lightweight, low-inertia end effectors because the main actuators, the skeletal muscles, can be kept at rest.
Muscles connect to the distal bone only via tendons, which have a negligible weight. In this way, two passive safety aspects, which minimize the head injury criterion, are intrinsic to the anthropomimetic musculoskeletal architecture: compliance & minimal moving mass.
Bioinspired approaches on the controller side are simulated biological neural networks, because the human brain and central nervous system are the most relevant reference for natural control of musculoskeletal limbs. Neural control is the most elegant, versatile, and energy-efficient way to use musculoskeletal systems.
what it takes to make a robot strong
making science happen
The team of over 100 students, doctoral candidates and graduates of the Technical University of Munich brings together experts from a wide variety of disciplines. Together with a network of scientists all over the world you have been working for years on the development of the humanoid robot. The Royal Institute of Technology in Stockholm (neuroprosthetics), the Chinese University of Hong Kong (algorithms for controlling the robot), Oxford University (loading of artificial tendons during their growth) and of course TUM (robotics & real-time systems, product development methods) are permanent cooperation partners.
Why you need to have MUSC
a robotics toolkit that fits its user needs
Musc crab-style legs
Minimal 2DoF setup.
6 muscles, 3 DoF shoulder
Full-size humanoid robot.
keeping it open – it’s the only way for research
The development of Roboy (mechanic and software) is conducted open source. This means that all expertise, ideas, and inventions do not belong to one specific entity, and everyone will have the chance to advance Roboy’s technology.
All of the code and CAD files are freely available on GitHub under a very permissive license (BSD 3.0 and CC-BY 4.0).
The parts required to build Roboy are kept in our sponsored aligni.com instance at roboy.open-aligni.com.
The documentation of the work of the current individual teams is available on our team development space.
As the project is under heavy development, if you would like to have access, learn how to build your own, or contribute please contact .
Think Musc could be valuable for your research? Get in contact with us, we’re happy to help building or build for you!
The Roboy Dialog System
to say and what to say, that is the question!
The Roboy Dialog System (RDS) is a sophisticated software module representing the cognitive capabilities of the humanoid anthropomimetic robot Roboy. The goal of the project is to implement dialog routines and knowledge extraction for a realistic human-like conversation flow which is achieved by utilizing various behaviour models represented by the State Machine (RDSM) finite automaton defined via a certain Roboy Personality description (file). Within the particular conversation flow stages, the behavioral variability is obtained by extending and redefining the common RDSM State to produce a certain social interaction. The RDMS State both as actor and as reactor regarding the internally formulated output and externally acquired input.
The architecture of the Roboy Dialog allows for flexible configuration and achieving the dialog flow tight to the user needs. Inherently it is designed for the open-domain conversation, which allows to handle unpredicted user inputs. The system can be deployed in multiple scenarios, including
In addition, Roboy Dialog System is capable of sending facial expression commands or Roboy movement action calls.
For the general knowledge information access, the interface to the DBpedia is implemented, as well to the internal graph-based memory, that allows to persistently store information across dialog sessions.
We ❤️ Robotics
that’s why we do the research reviews
Note: We are reviewing these papers, and we were not part of the research or the team which authored this paper.