The company owned by Google "DeepMind" known for its developments in the field of artificial intelligence and the construction of neural networks capable of playing computer games at the human level, recently announced the engine to simulate physical processes MuJoCo (Multi-Joint Dynamics with Contact).
The motor tIt aims to model articulated structures that interact with the environment, and it is used for simulation in the development of robots and artificial intelligence systems, in a stage prior to the implementation of the technology developed in the form of a finished device.
MuJoCo reaches a sweet spot with its contact model, which accurately and efficiently captures the salient characteristics of objects in contact. Like other rigid body simulators, it avoids fine detail deformation at the contact site and often runs much faster than in real time. Unlike other simulators, MuJoCo solves contact forces using the convex Gaussian principle.
Convexity ensures unique solutions and well-defined inverse dynamics. The model is also flexible, providing multiple parameters that can be adjusted to approximate a wide range of contact phenomena.
The code is written in C / C ++ and it will be released under the Apache 2.0 license and it will have support for Linux, Windows and macOS platforms. The opening work of all source codes associated with the project is scheduled to be completed in 2022, after which MuJoCo will move to an open development model, which implies the possibility of participation in the development of community representatives.
About MuJoCo
MuJoCo is a library with a general purpose physics simulation engine what se can be used in the research and development of robots, biomechanical devices and machine learning systems, as well as in the creation of graphics, animation and computer games. The simulation engine is optimized for maximum performance and enables object manipulation at a low level, while providing high precision and rich simulation capabilities.
Because many simulators were initially designed for purposes like games and movies, they sometimes take shortcuts that prioritize stability over precision. For example, they can ignore gyroscopic forces or directly modify speeds. This can be particularly damaging in the context of optimization: As artist and researcher Karl Sims first observed, an optimization agent can quickly discover and exploit these deviations from reality.
In contrast, MuJoCo is a second-order continuous time simulator that implements the full equations of motion. Familiar but not trivial physical phenomena like Newton's Cradle, as well as unintuitive ones like the Dzhanibekov effect, emerge naturally. Ultimately, MuJoCo sticks closely to the equations that govern our world.
Models are defined using the XML-based MJCF scene description language compiled with a dedicated optimization compiler. In addition to MJCF, the engine supports uploading of files in the Unified Robot Description Format (URDF). MuJoCo also provides a graphical interface for interactive 3D visualization of the simulation process and the representation of results using OpenGL.
Of its key characteristics, the following stand out:
- Simulation in generalized coordinates, eliminating joint breaks.
- Reverse dynamics, detectable even when there is contact.
- Use of convex programming for the unified formulation of constraints in continuous time.
- Ability to set various restrictions, including soft touch and dry friction.
- Simulation of particle systems, tissues, ropes and soft objects.
- Actuators (actuators), including motors, cylinders, muscles, tendons, and crank mechanisms.
- Resolution programs based on Newton's methods, conjugate gradient and Gauss-Seidel.
- Possibility of using pyramidal or elliptical friction cones.
- Use of a selection of Euler or Runge-Kutta numerical integration methods.
- Discretization and multiprocess approximation by the finite difference method.
Finally, if you are interested in knowing more about it, you can consult the details In the following link.