Download at: https://github.com/ReactRobotics/DogBotV4
React Robotics are creators of the DogBot quadruped. This is a robot primarily aimed at Machine Learning research, so in the real world it is optimised for fast communication (EtherCAT), high-torque (60Nm nominal, 120Nm peak) and low weight (20kg). Torque to weight ratio is therefore comparable to MIT Cheetah 3 and Mini Cheetah - while these robots have been demonstrated jumping on tables and doing backflips, we haven't tried any of those things yet, but in theory they should be possible. A simulation is a good place to start though.
Current status of the robot hardware is in late stages of a pre-production prototype, and will be released soon - stay tuned. Contact us if you are doing really interesting research and want a use a quadruped.
So far we have NOT tested the URDF in MuJoCo, so we appreciate your feedback on how this works in MuJoCo environment. The URDF has been tested in Gazebo and using the DART sim under OpenSceneGraph - it works here fine, apart from the usual contact friction discrepancies. Joint torques - the Effort parameter - are currently set at 60Nm, so you can increase these to 120Nm if needed.
We are happy to support you in any way we can. Contact us on the MuJoCo forum with questions if you have any, or raise an Issue on GitHub page.
If you create any cool videos then please post a link on my page: http://www.mujoco.org/forum/index.php?members/nic-greenway.2056/ or tweet them @reactrobotics.
Image below: Real robot
Image below: Simulated robot