D'Claw is platform introduced by project-ROBEL (RObotics BEnchmarks for Learning) for studying and benchmarking dexterous manipulation. It's a nine degree of freedom platform that consists of three identical fingers mounted symmetrically on a circular laser cut base. The finger tips are 3D printed parts. The base can be fixed to a stationary position, or mounted to a portable frame (called D'Lantern).
The field of reinforcement learning is growing exponentially, but cutting-edge research requires time, money, and infrastructure. This lack of accessibility in research is slowing progress in the field, and makes reproducing or comparing results almost impossible.
To tackle this problem, we're excited to introduce ROBEL (RObotics BEnchmarks for Learning): a collection of affordable, reliable hardware designs for studying dexterous manipulation and locomotion on real-world hardware. With simple assembly instructions, detailed simulations, and all open-sourced software, we hope to open up the field of Reinforcement Learning to everyone and accelerate progress around the world
On-hardware training with DAPG effectively learns to turn a flexible objects. We observe manipulation targeting the center of the valve where there is more rigidity. D'Claw is robust to sustain on-hardware training facilitating results on hard to simulate tasks.
A Sim2Real policy trained via Natural Policy Gradient on MuJoCo simulation with object perturbations (amongst others) being tested on hardware. We observe fingers collaborating to together to resist the external disturbances
A Sim2Real policy trained via Natural Policy Gradient on MuJoCo simulation with external perturbations (amongst others) being tested on hardware. We observe that free fingers fill in for the missing finger
On-hardware training with distributed version of EC-SAC leaning to turn multiple objects to arbitrary angles in conjunction by sharing experiences. Five tasks only need twice the amount of experience of single tasks, thanks to the multi-task formulation. In the video we observe five D'Claws turning different objects to 180 degree (picked for visual effectiveness, actual policy can turn to any angle)
DYNAMIXEL XM430-W210-R x 11
DYNAMIXEL U2D2 x 1
6 Port RX/EX Power Hub x 1
HN12-I101 SET x 2
FR12-S102K SET x 6
FR12-H101K SET x 9
HN12-I101 SET x 2
Nuts & Bolts for assembly
Please Note: D'Claw will require additional 3D printed / laser cut pieces to assemble, for more information please refer to the D'Claw Parts and Assembly Files
Getting Started Guide
Google's ROBEL D'Claw evolved from earlier designs Vikash Kumar developed at the Universities of Washington and Berkeley. Multiple people across organizations have contributed towards the ROBEL projects. We thank our co-authors Henry Zhu (UC Berkeley), Kristian Hartikainen (UC Berkeley), Abhishek Gupta (UC Berkeley) and Sergey Levine (Google and UC Berkeley) for their contributions and extensive feedback throughout the project. We would like to acknowledge Matt Neiss (Google) and Chad Richards (Google) for their significant contribution to the platform designs. We would also like to thank Aravind Rajeshwaran (U-Washington), Emo Todorov (U-Washington), and Vincent Vanhoucke (Google) for their helpful discussions and comments throughout the project.