SpotMicroAI - SpotMini-Clone with S

Introduction

I started this Project a few weeks ago, so it's still at the very beginning and HEAVILY work in progress. ;)

Goal of this Project is to have a tiny working "Clone" of Boston-Dynamic's SpotMini and to use ML/RL to do/support leg/body-motion.

It is based on existing OpenSource-Projects and uses a 3D printed base and affordable Hardware to enable other People to build their own Robots and play around with Kinematics and Simulation to understand how all this works. So that one day there will be affordable four-legged OpenSource Robotic-Platforms everyone can build and put "logic" on top of.

So everyone of you is invited to contribute to this Project.

see https://github.com/FlorianWilk/SpotMicroAI

Step 1 — the physical Robot

The physical Robot is based on KDY's awesome Thingiverse-Project: https://www.thingiverse.com/thing:3445283

I replaced the original ArduinoMega, which controls the Servos with an NVIDIA Jetson Nano + a 16 Channel PCA9685 I2C-Servo Driver.
Then it has 4 HC-SR04 SonarSensors, two in the Front and two in the Bottom. It also has a IMU MPU-6050.
And i added a RaspberryPi-Cam to the Front which will be used by the Jetson to do simple ObjectDetection (maybe MonoSLAM?).

Here is how it looked after the first sanding and priming and sanding with the first Hardware-Setup (ArduinoMega)

Step 2 — the Simulator

I needed a Simulation for several reasons. I currently use PyBullet to do Simulation for kinematics and RL in the future.
There is no real RL-Training in there yet, but that's my main reason for this Project. I want to understand how ML/RL can support a Physical Robot. Maybe it even learns how to walk some day. See https://arxiv.org/pdf/1804.10332.pdf for Inspiration ;)

Step 3 — RL Training

no RL yet :( i am still working on Kinematics

Flag this post

SpotMicroAI - SpotMini-Clone with Simulation-Environment and RL

Step 1 — the physical Robot

Step 2 — the Simulator

Step 3 — RL Training

Thanks for helping to keep our community civil!