Inverse Recurrent Models - An Application Scenario for Many-Joint Robot Arm Control

Sebastian Otte, Adrian Zwiener, Richard Hanten and Andreas Zell
This project investigates inverse recurrent forward models for many-joint robot arm control. First, Recurrent Neural Networks (RNNs) are trained to predict arm poses. Due their recurrence the RNNs naturally match the repetitive character of computing kinematic forward chains. We demonstrate that the trained RNNs are well suited to gain inverse kinematics robustly and precisely using Back-Propagation Trough Time even for complex robot arms with up to 40 universal joints with 120 articulated degrees of freedom and under difficult conditions. The concept is additionally proven on a real robot arm. The presented results are promising and reveal a novel perspective to neural robotic control.

Supplementary video material for our ICANN 2016 submission

The following videos show movement trajectories of different robot arms converging towards random poses. The blue coordinate system represents the target pose, whereas the red one represents the reference frame of the end-effector tip. The trajectories directly reflect the iteration loop - following the gradient from the recurrent forward model. Note that for producing the videos the maximum step length (in configuration space) was reduced in order to yield a slower motion trajectories at 25 fps.

Moving towards random poses (5 joints)

Moving towards random poses (10 joints)

Moving towards random poses (20 joints)

Moving towards random poses (40 joints)

The video below demonstrates that our approach also works for real robot arms. The robot arm performs an 'eight' trajectory. For all trajectory points (in Euclidean space) the inverse kinematics were computed successively via the inverted recurrent neural network.

Crustcrawler arm performing 'eight' trajectory

References

[1] Adrian Zwiener, Christian Geckeler, and Andreas Zell. Contact point localization for articulated manipulators with proprioceptive sensors and machine learning. In 2018 IEEE International Conference on Robotics and Automation (ICRA), pages 323--329, May 2018. [ DOI ]
[2] Adrian Zwiener, Sebastian Otte, Richard Hanten, and Andreas Zell. Configuration depending crosstalk torque calibration for robotic manipulators with deep neural regression models. In Intelligent Autonomous Systems (IAS), The 15th International Conference on, Baden-Baden, Germany, 2018. (Accepted for publication). [ details ]
[3] Sebastian Otte, Adrian Zwiener, and Martin V Butz. Inherently constraint-aware control of many-joint robot arms with inverse recurrent models. In International Conference on Artificial Neural Networks, pages 262--270. Springer, 2017.
[4] Sebastian Otte, Zwiener Adrian, Richard Hanten, and Andreas Zell. Inverse Recurrent Models -- An Application Scenario for Many-Joint Robot Arm Control. In International Conference on Artificial Neural Networks (ICANN), pages 149--157, Barcelona, Spain, September 2016. Springer.