#A bit of a suggestion for Neurodog

1 messages · Page 1 of 1 (latest)

steady talon
#

This is swarm from China, I've seen clips of Ellie having trouble with the Neurodog. And I'm trying my best to be helpful by giving a bit of suggestions. I major in Robotics Engineering in university.

The main problem is the leg design. Back in serveral months ago when I see the photo of ellie holding the first leg of Neurodog I felt it a bit weird. Ellie put two motors in the root of the leg, and put the motor that powers knee joint staright at the knee. This design is common in the easily-found cheapest robot arm on amazon (or temu or anything). This design is approximately fine for robot arm since it doesn't have to bear much load for the most time. If you have ever closely inspected Go2 from Unitree, you will find that the calf(or lower part of the leg) is connected to the motor placed in the root of the leg with a plastic rod. This design can effectively decrease the rotational inertia of the leg, while avoiding the space restraint of knee joint on the motor. I suggest replacing the leg design with something similar to this.

The second problem is the selection on the motor. Given the photo ellie posted on twitter on 12/2024, I'm sure she picked servo as the motor powering the leg. Ellie mentioned that the motor reacts slowly to the command that she gives. I think servo is the souce of this problem. Servo has a relatively small torque as well. I suggest replacing the motor with planetary gear motor with integrated optical, hall effect, or absolute encoder. it should provide a higher maximum speed, much more maximum torque, and a much shorter reaction time while maintaining the perception of the angular postion of the motor. In addition, the motor closer to the root of the thigh should generally be more powerful than the lower ones, but ellie made them same, this could cause problems too.

#

If Ellie wants to upgrade it, the main obstacle should be the space restraint at the root of the leg. Ellie did mentioned that her mechanical design is based upon the selection of the motor. Thus, there might not be enough space to contain a more powerful motor. But I believe this can be solved.

#

But at the end of the day, it is reasonable that Ellie designed Neurodog like this. This design has the easiest development difficulty in hardware and software both. It's a prototype afterall.

steady talon
#

But not gonna lie, the fastest way to make a Neurodog movable and talkive, is probably to buy a Go2 of Unitree or Spot from BostonDynamics. Then remove its front plate, replace it with a phone, ask Anny(or someone else) to decorate it nicely, and develop it purely by software. I've heard that there is going to be a meeting IRL in UK in about a month. I think ellie mentioned she wanted to set the debut of Neurodog by then. I'm not exactly sure when is it, nor do I know whether in this way will we do it in time.

grand gazelle
#

I don't understand anything about it, but I hope Vedal sees it and passes it on to Ellie. This looks really useful YES

steady talon
#

Picture reference for the structure of the leg of Go2

left yew
#

Consider reposting this in Ellie's discord, perhaps in #engineering-watercooler or #ellie-suggestions

#

I've sent you an invite link to their discord if you want to join. If you post it there, Ellie will certainly see it.

left yew
left yew
#

I failed to find the stream 😔, but it was certainly on a dev stream before 20 August 2024. People were suggesting that Vedal buy the Unitree robodog, because it was only around $3k. Vedal then said that those robdogs weren't programmable. He said he would need to get the educational Unitree robodog that came with an SDK, which was around $10k.
Vedal wants to implement all the computer vision, object detection, pathing etc. himself (whilst the robodog handles uprighting. movement).

#

can effectively decrease the rotational inertia of the leg
Ahhh, so less force is needed to rotate the leg (and your motors effectively become more powerful)? Makes sense!