Why DeepMind acquired this robotics startup
Earlier this week, Alphabet-owned DeepMind acquired a physical simulation platform MuJoCo, which stands for Multi-Joint Dynamics with Contact.
After the acquisition, the DeepMind Robotics Simulation team, which used MuJoCo in the past, plans to fully open the platform in 2022 and make it freely available to anyone to support research everywhere.
MuJoCo was first developed by Emo Todorov for Roboti and was available as a commercial product from 2015 to 2021. After DeepMind acquired MuJoCo, it makes it available to everyone for free. However, details of the financial transactions have not yet been disclosed.
After acquisition, Roboti will continue to support existing paid licenses until they expire. In addition, the legacy version of MuJoCo (versions 2.0 and earlier) will remain available for To download, with a free activation key dossier, valid until October 2031.
What is MuJoCo?
MuJoCo is a physics engine that aims to facilitate research and development in robotics, graphics, biomechanics, animation and other fields requiring fast and accurate simulation. It is one of the first full-featured simulators designed from the ground up for model-based optimization, especially through contacts.
The platform enables computation-intensive techniques such as optimal control, physically consistent state estimation, system identification and design of automated mechanisms to be scaled and applied to applications. complex dynamic systems in behaviors rich in contacts. Additionally, it has more traditional applications such as testing and validating control schemes prior to deployment to physical robots, interactive science visualization, virtual environments, animation, and games.
How is MuJoCo different?
DeepMind MuJoCo is not alone. Other simulation platforms include Facebook’s Habitat 2.0 and ManipulaTHOR from AI2. What sets them apart, however, is their contact model, which accurately and efficiently captures the main characteristics of objects in contact. Like other rigid body simulators, it avoids fine detail of deformations at the contact site and often operates much faster than real time.
âUnlike other simulators, MuJoCo solves contact forces using the Gauss principleSaid the DeepMind Robotics Simulation team. Convexity guarantees unique solutions and well-defined reverse dynamics. Additionally, the model is flexible, providing multiple parameters that are tuned to approximate a wide range of contact phenomena.
Additionally, the DeepMind team said their platform is based on real physics and doesn’t take any shortcuts. According to them, many simulations were originally designed for purposes like games and movies; they sometimes take shortcuts that favor stability over precision. For example, they can ignore gyroscopic forces or directly change speeds.
This, as part of the optimization, can be particularly harmful. In contrast, MuJoCo is a second order continuous time simulator, implementing the full equations of motion. In other words, MuJoCo adheres tightly to the equations that govern our world – non-trivial physical phenomena like Newton’s cradle, and those not very intuitive like the Djanibekov effect, happens naturally.
The team also said that MuJoCo’s mid-engine is written in pure C, which makes it easily portable across various architectures. In addition to this, the platform also provides quick and convenient calculations of commonly used quantities, such as kinematic Jacobians and inertia matrices.
MuJoCo offers powerful scene descriptions. It uses cascading defaults – avoiding multiple repeated values ââ- and contains elements for real robotic components such as tendons, actuators, equality constraints, motion capture markers, and sensors. Soon he plans to include the standardization of MJCF as an open format to extend its utility beyond the MuJoCo ecosystem.
On top of that, MuJoCo includes two powerful features that support musculoskeletal models of humans and animals. It captures the complexity of biological muscles, including activation states and force-length-speed curves.
DeepMind Slaying Robotics
DeepMind has invested heavily in robotics research. Recently he introduced RGB-Stacking, a new benchmark for vision-based robotic manipulation.
The recent acquisition comes at a time when there is a dearth of data in robotics research. This is one of the reasons why OpenAI, DeepMind’s big rival, has shut down its robotic arm indefinitely. But that doesn’t stop DeepMind, as its teams are trying to work around this lack of data with a technique called sim-to-real, in a big way.
Now, with the acquisition of MuJoCo, the open source library appears to be a smooth evolution for the company, and will surely benefit the robotics ecosystem as a whole.
Subscribe to our newsletter
Receive the latest updates and relevant offers by sharing your email.