im about to finish the build for the Bolt bipedal Robot.
while im very exited to work on my own software implementation, I would like to start with the reactive_planner Demo to confirm everything is functional.
Currently im very much into the hardware setup, and therefore haven’t spent to much time of understanding the full software workflow and dependency’s.
The Simulation is already up and running without any issue.
However, when watching all the Videos published by ORDI, I recognize that usually the robots have the Vicon Makers installed.
Is that a requirement for the reactive_planner Demo to run ? or does it come with its own state estimator based on forward kinematics and IMU ?
Also i assume that the Bolt Codebase is utilizing the same Low Level Ethernet connection to the Masterboard like known from the Solo implementation.
Any recommendation whether I should go for the Ethernet or Wifi connection , any Pros and Cons ?
I really appreciate the great effort of making such a beautiful project open to the public
Yes, for controlling the robot you would need a state estimation module. If you have access to the Vicon system, it simply gives you the state of the robot base, but if you do not have that you would need a state estimation module. Our state estimation codes are not still open-source (we’ll open-source it soon), but there are other available software you could use for instance GitHub - ori-drs/pronto: Pronto - Legged Robot State Estimator - libraries, ROS wrapper and messages .
For the communication, yes it’s the same Ethernet connection like Solo. The Wifi module is not very reliable at the moment, so I would not recommend it.
When in comes to the State estimator, can you provide any performance parameter like frequency and precision that is feasible for the demo code ?
I assume the Vicon is very precise compared to a onboard state estimator ?
Also what are the exact parameter that are effectively used from the Vicon, is it just X/Y/Z Space position of the body ?
I cannot give you a number for precision, the more precise the better. But you need the states at 1 KHz since your whole body controller is running at 1 KHz. Normally the frequency of Vicon is less but that is not a problem. About the precision of Vicon compared to state estimator, Vicon is better in terms of base pos estimation, but its velocity is noisy and needs filtering. The state estimator gives very clean velocity estimates but the pos drifts over time.
The most important base quantities that are required for control are base linear and angular velocity, base orientation, and base height position (base x and y are not important for control).
is there any chance you publish your State Estimator implementation soon ?
apart from the state estimator, is there something else I need to take care of to run the reactive planner demo on a real Bolt ?
We’ll do it in the next days, I’ll let you know once it’s done. The only thing is that the estimator is for Solo, but with slight changes it should work for the Bolt as well.
I don’t think you would need anything else to make the reactive controller work, you’d only need to be patient and consistent with the experiments
