Comparison between experiments and two simulators with two different low-level trajectories tracking methods

The objective of this research is that we realized there always exists a gap between simulation and real experiments. When you have a lot of parameters in your controller, adjusting them is time consuming. In most cases, even if you have gotten parameters worked in simulation, you need to change them a lot in the real experiment. By this research, we want to find which simulator is closer to reality so that the parameters worked in the simulation will be more meaningful. Also, we want to analyze the reason that caused such a difference.

More details...

The final goal of our research is comparing the performance of different simulators and different low-level control methods in biped robot Cassie. To achieve this final goal, I divided it to some small steps and realized them:

  • Implemented a walking controller in C++ and tested it in simulation:

    For the low-level control, both PD control and Passivity control can be used. Here is the video for Cassie walking in simulation.




  • Combined invariant extended Kalman filter with the controller in ROS to decrease the error in velocity tracking:

    The red line shows the velocity obtained from inEKF and the blue line shows the true velocity gotten from the simulator. The velocity track is really good from inEKF!




  • Tested Cassie in really experiment:

    After some adjustments, we are able to make Cassie walk in real experiment with our C++ controller!






Acknowledgement

Thank you to Professor Jessy Grizzle and Grant Gibson for guidance!
Thanks for the help from Yukai Gong and Bruce JK Huang!