Google-owned company DeepMind, famous for its developments in the field of artificial intelligence and the construction of neural networks capable of playing computer games at the human level, announced the discovery of an engine for simulating physical processes MuJoCo (Multi-Joint dynamics with Contact). The engine is aimed at modeling articulated structures interacting with the environment, and is used for simulation in the development of robots and artificial intelligence systems, at the stage before the implementation of the developed technology in the form of a finished device.
The code is written in C/C++ and will be published under the Apache 2.0 license. Linux, Windows and macOS platforms are supported. Open-source work on all of the project's content is expected to be completed in 2022, after which MuJoCo will move to an open development model that allows community members to participate in the development.
MuJoCo is a library that implements a general-purpose physical process simulation engine that 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 allows low-level object manipulation while providing high accuracy and rich simulation capabilities.
Models are defined using the MJCF scene description language, which is based on XML and compiled using a special optimizing compiler. In addition to MJCF, the engine supports loading files in the universal URDF (Unified Robot Description Format). MuJoCo also provides a GUI for interactive 3D visualization of the simulation process and rendering of the results using OpenGL.
Key features:
- Simulation in generalized coordinates, excluding the violation of joints.
- Reverse dynamics, determined even in the presence of contact.
- Using convex programming for a unified formulation of constraints in continuous time.
- Ability to set various constraints, including soft touch and dry friction.
- Simulation of particle systems, fabrics, ropes and soft objects.
- Executive elements (actuators), including motors, cylinders, muscles, tendons and crank mechanisms.
- Solvers based on Newton's methods, conjugate gradients and Gauss-Seidel.
- Possibility of using pyramidal or elliptical friction cones.
- Using the choice of numerical integration methods of Euler or Runge-Kutta.
- Multithreaded discretization and approximation by the method of finite differences.
Source: opennet.ru