• Mapping using Rtabmap with D455 and 2d lidar
• Navigation using move_base

Make 3

• uses old dynamixel RX28 servos.
• 18 dof
• RPLidar A1 2d Lidar
• Jetson Xavier NX
• Intel Realsense d455 depth camera
• runs openshc ROS package (https://github.com/csiro-robotics/syropod_highlevel_controller)
• tip ground contact sensors (tactile switches) using rosserial_arduino
• wrote a ROS node that publishes the tip sensor data in a format openshc understands.
• want to make it into a decent rough terrain crawler
• servos seem to reach thermal limit quickly - added fans to motors
• 2d and 3d slam using a 2d lidar and a depth camera with hector mapping and rtabmap packages

Videos

Indoor 3d SLAM with d455 and exploring various sensor data

Here I manually control the hexapod with a ps4 controller around the room. ROS master node is running on the bot on a Jetson NX. It has a 2D lidar and a Realsense depth camera. It runs the openshc package for hexapod control, and rtabmap for 3D SLAM. Visualization is done on a desktop machine on the same network.

Long video of testing the admittance controller

Short screen capture of exploring variables on rviz/rqt

Outdoor flat forest floor traverse test

Some Pictures

Side by side with Make 2

Assembled

Close up

Visualization of topic data from openshc with rqt and rviz

Added fans

Silicon anti slip cap (and the mould)

Make 2

• uses dynamixel XL430 servos.
• 24 dof
• the same control system as Make 1.
• runs on a Jetson Xavier NX.
• has a 2d lidar ~ 10Hz
• has a stereo depth camera (realsense D435i)
• experimented with running SLAM stack on ros using the two sensors: https://www.youtube.com/watch?v=-N8fmfxq5lY
• played around with hector slam for 2d lidar, rtabmap, VINS-Fusion (https://github.com/HKUST-Aerial-Robotics/VINS-Fusion)
• got familar with the high level organization of the ROS stack.
• later removed one dof from each leg for faster IK computation.
• control software capability (even ground mode): https://www.youtube.com/watch?v=li7Eii5mxpQ

4 DoF Legs

3 DoF Legs

Make 1

Inspired by Smallp Tsai’s hexapod v2.1 video

Drawings

Notes

• used a raspberry pi zero sized board
• has an IMU
• 24 dof
• control software written in python using numpy, scipy, servo controller library.
• 4dof legs -> no analytical solution for inverse kinematics. Implemented an iterative method for 4dof leg IK solution. Also can use an optimizer from scipy.optimize.
• basics of the endeffector position based control algorithm code in this jupyter notebook
• power system is janky - sometimes when the servos draw too much power at once, the sbc reboots.
• learned to use Fusion360 to design parts needed using a 3D printer.
• parts:
  • banana pi zero
  • 7x buck converter
  • GY210 IMU
  • 2s lipo
  • 24x towerpro MG90B servos
  • 2x 12ch pwm controller board
• one buck converter provides 5v to the sbc
• the servo controller boards get power from sbc
• four servos (one leg) get power from one buck converter
• sbc, servo controller boards, and the IMU talk through i2c

Pictures

Videos

3 DoF Legs

4 DoF Legs

BoM

• 6x Dynamixel XL430 (~300 EUR)
• 2x Dynamixel 2XL430 (~200 EUR)
• 1x RPLIDAR A1 (~100 EUR)
• 1x Intel Realsense D4xx Camera (200 ~ 300 EUR)
• Jetson Nano / Xavier NX / Raspberry Pi (50 ~ 400 EUR)
• 3d printed parts
• 12V DC step down converter (< 10 EUR)
• (5V DC step down converter, necessary for Jetson Nano & Raspberry Pi) (< 10 EUR)
• screws (~ 30 EUR)
• 4s Lipo Battery

With the choice of the board	Total cost around
Jetson Xavier NX	$1250
Jetson Nano	$1000
Raspberry Pi	$920

Arm configurations

Navigation 1

• High camera position, better position of the camera for 3d SLAM
• More camera shaking during movement from backlash and tolerances

Navigation 2

• Lower camera position, worse position of the camera for 3d SLAM
• Less camera shaking during movement from backlash and tolerances