A small 12-DoF printable legged robot using quasi-direct drive brushless actuators
Already have an account? Log in.
To make the experience fit your profile, pick a username and tell us what interests you.
Build instructions in video :
Mechanical design viewer : https://autode.sk/3qW5eA4
ESC Firmware :
Actuators based on project: "Ultra low-cost controller for brushless servo based on the B-G431B-ESC1 Discovery kit by ST" : https://hackaday.io/project/177578-b-g431b-esc-brushless-servo-controller
Custom Firmware for ST Discovery Kit B-G431B-ESC1 may be downloaded from git : https://github.com/pat92fr/BrushlessServoController
ROS node :
Patrick.
Testing a foot trajectory at high speed using Bezier curve.
I was looking for a smooth swing phase trajectory for my DIY legged robot, that suits a large range of robot velocity (from 0 to 1 m/s and more). This trajectory shall be computed in real time, and should minimize foot velocity and acceleration, and joint velocity and acceleration, in order to get a good tracking and minimal ground impact force at the begining of the stance phase.
The Bezier curves solution is not yet perfect! There is still a position overshoot at the begining of the stance phase, that could cause a ground impact at the touch down. Foot position tracking at quite a high velocity is not too bad though, and depends on the foot trajectory curve and also on the brushless motor controler setup (Kp and Kd of the position control loop, Kp and Ki of the FOC current loop, FOC implementation). The PID has been set for the a robot walking on the ground, that maybe not be optimal when the robot is not in contact with the ground. Well, that require further analysis, but those first experiments are interesting.
Git : https://github.com/pat92fr/FootTrajectoryPlanner
Patrick.
See. Instructions v0.01.pdf in file section.
Any advice about the below GIM4305 motor compared to the MAD5008 motor? Thanks!
https://www.aliexpress.com/item/1005003433829950.html?spm=a2g0o.productlist.0.0.21ad27cdOMVPLQ&algo_pvid=c43770ea-4ea3-40b9-9edc-66e7a0e05210&algo_exp_id=c43770ea-4ea3-40b9-9edc-66e7a0e05210-1&pdp_ext_f=%7B%22sku_id%22%3A%2212000025769288175%22%7D&pdp_npi=2%40dis%21USD%21%2184.9%21%21%21%21%21%400b86d81616535750300914968e76fa%2112000025769288175%21sea&gatewayAdapt=4itemAdapt
Hey, I have seen thismotor/reducer too. For a 3.5kg robot , I think these motors are suitable for such a quadruped. Test must be completed ti check the strength of reducer (impact resistant gears).
Patrick.
Awesome project!
Any advice if hope to reduce the weight from 5.5Kg to 3.5kg? Thanks!
Hello Afreez Gan,
An overall weight of 3.5kg was my target during mechanical design.
Note : The weight of the 12 motors is about 1.7kg.
First, the Up Board is quite heavy, and such a powerfull SBC is not required for this robot. Replaced by a RPi+CAN HAT or a even smaller computer (M5Stack) is possible. Expected weight reduction : 400g.
Second, wiring is quite heavy due to the reuse of COTS ESC. There are a lot of cables, connectors (dupont,xt30/xt60), stand-alone magnetic encoder boards,
and even intermediate boards for power and CAN distribution. By redesigning the FOC controller, with daisy chain for both CAN and power, the expected weight reduction should be about 400g.
Third, reducers and legs are quite heavy too (800g for the 4 legs for instance). By replacing FDM parts (legs, gears, reducers, frame...) by injected/molded plastic parts, and by redesigning parts with a higher level of optimisation (more simulation of forces and deformation), I think that an noticeable weight reduction is possible. By te way, molded plastic parts allow suppression of a few M3 to M5 screws.
By using materials like nylon and PTFE, some ball bearings may be suppressed too.
A weight of 4kg may be possible, and the performances of the robot will be greatly improved.
Patrick.
Love your professional and detailed advice, so great! Thanks a lot!
Comments about robot design and programming are welcome. Thank you!
pat92fr
Tim Wilkinson
Nyeli Kratz
ssla-couk
ensafatef
why did you design the leg like a chain of link, not like mit cheetah.
What is the benefit for design like that.