Hello,
We are currently building the biped Bolt hardware. We have a couple of questions regarding the softwares, as there are many different packages on the Github…
Thanks!
Hi,
Here are some simple tests to check if the robot is working, but they are only for Solo12 and one-leg hopper but should be easy to extend it to Bolt:
Here are the demos for walking controller of Bolt:
For instance you can start with this reactive_planners/demo_reactive_planners_bolt_step_adjustment.py at master · machines-in-motion/reactive_planners · GitHub
Best,
Majid
how do you generate the reference traj?
when i do the simulation,i also found that the gournd reaction force is hard to handle it well,which cause sometimes the holonomic constraints may violated,Bolt will slip.
Thanks! I will take a look
Hello
I have installed reactive_planners successfully and run the Bolt demo without any error.
The Bolt in simulation always walk backwards,can not just step in place.
I am still reading the codes of reactive_planners and the related papers,no quite clear about how to modify the parameters to make Bolt just walk in place stably.
Can you tell me which parameters in demo_reactive_planners_bolt_step_adjustment.py can be used to control Bolt to walk in place, or forwards,or backwards?
Hi Alex,
Going always backwards could be a consequence of many things, e.g., the com is a bit behind the contact location in the standing configuration (legs are backward). There are many ways you can go around this issue, but the simplest and hackiest one is to command the robot a small forward velocity in the reactive stepping phase.
Best,
Majid
Thank you very much,it works.
As you’ve said,the initialize state plays an important role in stability.
If start from inperfect initialize state, is it possible to be stable after a few steps?
There is still another question i don’t know how reactive_planners how to solve it.
When i read other biped robot papers, like about Cassie or Mercury,those robot’s feet are designed as line feet to prevent yaw rotation.
But for Bolt,it seems that Bolt’s contact model is point-contact model.I have reading the code and related papers of reactive_planners for times,still can’t figure out how reactive_planners solve the yaw rotation problem.
Does there exist any papers or code which is solve the yaw rotation problem? Or this problem is actually not a problem for Bolt?
You are right in that we used a line foot to avoid yaw rotation (ankle of bolt is passive). Robots like Cassie use active ankle though, which on top of avoiding yaw rotation makes the balance of the robot a bit easier (it can statically balance itself while Bolt cannot). So for Bolt, there is no point in modeling the feet in simulation if we can find a hack to avoid yaw rotation, which is the thing we did, namely increasing the rotational friction at the contact point in simulation. In fact we tried to model the feet with a passive ankle but it was more harmful than useful (the robot sometimes lands on the end point of the foot that would lead to a quick fall).
Hi Majid:
I have running Bolt step adjust demo for many times, and i also found the real Bolt stepping video on youtube. It is really impressive to see real bolt walking so stable! The stance leg acts as fixed when impact occur,and the torso never shaking!! Even the ideal simulation still not perform so stable as you did for the real Bolt.
The papers do not mention how to deal with the impact and torso shaking problem,how do you solve these two problems for real Bolt?
Hi Alex, We didn’t do anything for impact, the compliance of the torque control system simply absorbs all the impact. Of course, that needs some gain turning to have all the time a fully damped system but there is no easy way to find the gains and you need to spend some time to figure it out.
Best,
Majid