Psst! Watch Phototropism, my new experiment.

Quadruped robot movement


I want to build a “four-legged spider” robot. It’s a fun project because 1) robots are cool and 2) I have very little experience with hardware, so I get to learn a lot of new things and be creative in new ways. Many areas are unknown unknowns to me — I don’t know what I don’t know.

I pick topics in a semi-random order and I build skills and create proof-of-concepts that I think will be useful for the final machine. So far I’ve played with soldering, 3D graphics, electronics, batteries, 3D printing, and remote control. Last week I was working on movement:

I was figuring out where to move legs, and in what order, to achieve a given move. It was a fun way to challenge spatial imagination. I built a repertoire of moves: get up, look around, walk forward, turn in place, walk sideways, walk in a circle (turn), lay down. Each of the moves is implemented as a separate routine that can be parametrized with different arguments: direction, angle, step length, duration, etc. I plan to use this work in the future application that will control the robot based on inputs from the user.

I focused on statically stable movement: only one leg is in the air at time. The other three form a triangle on the ground and the center of mass is inside it so that the robot won’t fall off. The movement can be stopped at any time and the robot will be stable.

In the future I want to experiment with dynamically stable movement so that the robot could move faster and with more grace. I didn’t work on it now because I don’t have the infrastructure to simulate gravitation and ground collisions in a way that won’t interfere with my movement logic yet — what you see is an animation not a simulation, there’s no gravity. I also don’t know the final dimensions or weight distribution of the robot because I haven’t built it yet. (But once I do I will be able to immediately test the moves you’ve seen above, so it made sense to prepare them first.)

I modeled the skeleton in Blender and implemented the animation as a Python plugin for Blender. In Python I set the location of the tip of each leg, location of the body, and body’s rotation. Blender’s inverse kinematics engine takes care of moving individual joints of the skeleton. When my plugin starts, it generates the full state for each frame of the animation. Blender reads the states frame by frame and updates the display. This way I can easily “crop” the animation to see only the fragment I am currently working on. I can also loop it or play it slower or faster without touching the code responsible for the movement itself.

That’s all for today. See you in the future in another robot post. ✌️🤖

Robotics series