4DOF Hexapod using ROS

First steps of assembled (mostly, Xtion and antenna mounts still missing) hexapod.
Some details:
Body is custom-milled aluminum

Each leg has 3 Dynamixel MX-64T servos for coxa, femur and tibia + AX-18A for tarsus

Dynamixels are controlled through USB2AX module

UM7 sensor for orientation control, connected through CHRobotics USB expansion board

Asus Xtion RGB-D sensor for mapping and obstacle avoidance

All kinematic calculations and processing of sensor data is done by on-board Intel NUC DC53427RKE

For now hexapod is remotely controlled with PS3 joystick, later some autonomous capabilities will be added

Software:
Ubuntu 14.04 + ROS Indigo
Pypot for controlling Dynamixels
ROS packages:
um7 for reading from UM7
openni for reading RGB-D from Xtion
depthimage_to_laserscan for converting RGB-D data to laserscan used gmapping
gmapping for SLAM
Movement is controlled by a self-made ROS node, written in Python and containing inverse-kinematics solver and gait controller.
Peculiarity of the gait controller is that there are no pre-defined gaits, all movements are calculated on the fly based on current position of thorax and limbs and target direction at this exact moment. New commands can be issued at any time and legs react instantly, without need to complete gait cycle. All legs are mostly independent of each other and only know whether immediate neighbors are landed or in swing. Whole gait forming is based on the restrictedness approach described in exhaustive detail in Michael Fielding’s thesis “Omnidirectional gait generating algorithm for hexapod robot”

Gait generation parameters have not yet been optimized, so eventually movement should be smoother.

Видео 4DOF Hexapod using ROS канала gbaitsur