About low-level servoing (or bypass ervoing) and possibile soulutions

[pos03052]

Summary

To enable real-time control, we have been transitioning from high-level control to low-level control over the past few weeks. We successfully implemented our own forward and inverse kinematics calculations, applied the actual URDF of our robot, and confirmed that it moves correctly.

Issue: Uncontrollable Joint Speed and Acceleration

However, we encountered a critical issue: in low-level control, As far as I know, velocity and acceleration of individual joints cannot be modified. As a result, when following a trajectory, the robot moves too fast.

My prevoius issue about low-level control modulation:

• https://github.com/Kinovarobotics/Kinova-kortex2_Gen3_G3L/issues/211

So, at high speeds, motion appears natural.
At low speeds, the robot reaches each target point at 1 kHz, causing a stop-and-go effect.
This makes smooth motion difficult, especially for precise and slow movements.

Potential Workarounds and Limitations

One possible workaround is to interpolate the trajectory or joint targets into smaller steps. This could partially mitigate the issue, it still does not provide smooth, general-purpose robot control. Instead, the behavior becomes more similar to a stepper motor, which is not ideal for many applications.

Questions and Alternative Solutions

Given these limitations, I would like to ask:

Are there any alternative solutions to this issue?
Would low-level bypass control be a viable way to resolve this?
From my understanding, low-level bypass control needs direct control of all seven actuators at 1 kHz, but I am unsure whether joint velocity and acceleration can be adjusted in this mode. If there is a way to achieve smooth, precise motion control at low speeds, I would appreciate any guidance on this.

[martinleroux]

Hi @pos03052 ,

In low-level position control, the robot provides a best effort (i.e. will maximize or adjust velocity and acceleration) to follow your requested trajectory. If you need the robot to go slower, you should be feeding smaller steps every millisecond. The average velocity of the robot will always be [step size] / [1 ms].

That being said, I am surprised you can perceive the robot stopping. Are you sure your communication rate is really 1kHz? If possible, you can send a code sample of your implementation to support@kinova.ca and we can sanity check it just in case.

Now to answer your questions:

1. Not really. Low-level control implies no planning, so the robot cannot have constraints on position, velocity and accelerations at the same time, unless all are within actuator limits and would result in the same displacement over 1ms - which is equivalent to only feeding position.
2. Low-level bypass offers no real additional functionality over low-level cyclic. In short, bypass is used when one wants to control actuators through 6/7 instances of ActuatorCyclic client instead of a single instance of BaseCyclic client. This could have minor benefits if you needed to shave microseconds in the communication part of your real-time loop, but I don't see any reason why it would help with your issue. In 99% of cases, I strongly recommend using Low-level cyclic due to the convenience of having a single interface through baseCyclic.
3. I have never heard of anyone having similar issues before. In theory, real-time control is best to get the robot to behave exactly as a user wants - as I said earlier I don't even understand how you can even perceive this. I think it may be easier to debug and help you if you could send a video showing the issue to our Support team, and then we could plan a live call to debug with you.

[pos03052]

Thank you for your detailed explanation. I understand what you mean.

I believe the communication cycle was 1 kHz, but I will first check a few things on my end.

I plan to verify the target trajectory, the inverse kinematics results (i.e., each joint angles), and the communication cycle. I suspect there might be some discontinuities in these values.
(This is because my understanding is that if the trajectory, joint angles, and control frequency are continuous at 1 kHz, the robot should move smoothly—even if the target trajectory moves very slowly.)

I don't think I will try low-level bypass for now as you recommended.

So, I will check for any discontinuities, and if I find any, I will correct them. If not, I will send you my code and a video.

Additionally, in low-level mode, can I expect the motor drive to generate a smooth trajectory under control of the maximum velocity and acceleration of the 7 actuators?

Thank you.

[martinleroux]

Yes. Be wary not too flirt too much with the limit though. If you run into Following Errors, it means your command is getting ahead of the actuator. The tolerance is pretty high (~5° if I remember correctly) but it can accumulate if you are just slightly over.