I wrote my own algorithm to generate gaits. It's simple but inconvenient to use. I'll try to organize and even write a paper for it. It may be too unprofessional to be accepted though.
To my knowledge, there are many professional simulation softwares used by robotics majored students and researchers. I'm waiting for those guys to train Nybble, and hopefully derive better results than I do.
sir i also asked you same question on your youtube channel and you told me something about man-learning so will you be able to elaborate and explain me
The algorithm uses tens of parameters for each gait. In short it uses inverse kinematics to create a mapping between joint angles and foot contacts. Design the trajectory of one leg then collaborate all four legs. Save the angles as animation frames and play it on Nybble. Imagine Nybble as a cheap electric keyboard compared with an expensive piano. It's up to you to play any melody on it.
When developing, I programed across different languages for my convenience. Being a short program, the physics simulation was done in my mind. Most of the tuning depends on my intuition and experience. It still takes several hours to fine tune one gait.
I hope people in serious robotics could teach us the right/professional way to generate gaits.
There are a lot of popular simulators, such as Webot. I haven't got a chance to learn them though.
Hi did you have time to collect all the data in order to explain your IK model? If not can you share some details I am really curious as some of the projects that uses IK models are not moving this fluent as Kitty. Really great and thank you for that!
My program was relatively simple but took many efforts to integrate and validate. I'm not going to share it because it's as informative as an academic paper and I also need to hold it for my business.
I would say that proper Intepolation plays an important role in making the motion smooth and reduce the overall calculation as well.
I'm interested in applying Machine Learning to develop walking gaits and other behaviors. This requires accurate 3D models for training in simulation. These must include dimensions, mass, rotational moments of inertia, as well as motor performance spec's. I don't necessarily need the exact shapes, just the relative position/orientation of joints and ground contacts (not trying to copy your molded parts).
Can you provide any of this information for Bittle?
yes, I have an early model of bittle that can be shared. It will be some time after the delivery. Currently we are full load on the final r&d and production.
Yeah, I would also like to use some RL approaches to improve Nybble and Bittle. Sharing what you have so far would be a really nice gesture.
Could you maybe also share some version of the simulator you where working on?
I wrote my own algorithm to generate gaits. It's simple but inconvenient to use. I'll try to organize and even write a paper for it. It may be too unprofessional to be accepted though.
To my knowledge, there are many professional simulation softwares used by robotics majored students and researchers. I'm waiting for those guys to train Nybble, and hopefully derive better results than I do.
sir i also asked you same question on your youtube channel and you told me something about man-learning so will you be able to elaborate and explain me
The algorithm uses tens of parameters for each gait. In short it uses inverse kinematics to create a mapping between joint angles and foot contacts. Design the trajectory of one leg then collaborate all four legs. Save the angles as animation frames and play it on Nybble. Imagine Nybble as a cheap electric keyboard compared with an expensive piano. It's up to you to play any melody on it.
When developing, I programed across different languages for my convenience. Being a short program, the physics simulation was done in my mind. Most of the tuning depends on my intuition and experience. It still takes several hours to fine tune one gait.
I hope people in serious robotics could teach us the right/professional way to generate gaits.
There are a lot of popular simulators, such as Webot. I haven't got a chance to learn them though.
Hi did you have time to collect all the data in order to explain your IK model? If not can you share some details I am really curious as some of the projects that uses IK models are not moving this fluent as Kitty. Really great and thank you for that!
Hi,
My program was relatively simple but took many efforts to integrate and validate. I'm not going to share it because it's as informative as an academic paper and I also need to hold it for my business.
I would say that proper Intepolation plays an important role in making the motion smooth and reduce the overall calculation as well.
Hi,
I just backed Bittle on Kickstarter.
I'm interested in applying Machine Learning to develop walking gaits and other behaviors. This requires accurate 3D models for training in simulation. These must include dimensions, mass, rotational moments of inertia, as well as motor performance spec's. I don't necessarily need the exact shapes, just the relative position/orientation of joints and ground contacts (not trying to copy your molded parts).
Can you provide any of this information for Bittle?
Thanks!
yes, I have an early model of bittle that can be shared. It will be some time after the delivery. Currently we are full load on the final r&d and production.
Yeah, I would also like to use some RL approaches to improve Nybble and Bittle. Sharing what you have so far would be a really nice gesture. Could you maybe also share some version of the simulator you where working on?
Good to hear. I was hoping I could work on this while waiting for delivery, but I understand you have a lot to do.
Thanks again.