DIY Quadruped Robot with Rough Terrain Traversal
Floor Location : M 125 N
Legged robots have a large variety of applications in fields such as space exploration and industrial maintenance. This is because legged robots, unlike wheeled or tracked robots, are able to explore and traverse a variety of terrain types, much like animals or people. My project is inspired by some of the pioneering technologies in legged robots, such as Spot from Boston Dynamics. The goal of this project was to build a cost effective robot that would be able to achieve autonomous movement through a variety of conditions. Such a robot would be able to explore and investigate potentially human-hazardous environments, or provide domestic assistance to people in need.
Throughout the development of this robot, the robot has gone through two major prototypes. Both prototypes started off being fully 3d modeled in Fusion 360, before being cut out on a CNC router and 3d printed. The robot has 4 legs, with 3 degrees of freedom each. The first prototype used a 16 channel servo controller to control the MG90s servos, though unfortunately the motors weren’t strong enough for the robot to walk more quickly. The second prototype uses serial communication to control Dynamixel ax-12a's, which were much stronger, provided torque feedback, and allowed the robot to walk more comfortably. As the AX12As communicate using half-duplex serial, a custom circuit board has been designed and created in order to allow the control system to communicate with the motors. A MPU-6050 gyroscope has also been used to calculate the robot’s orientation.
The entire robot is controlled and driven by a Raspberry Pi 4. The robot utilizes a self-developed control system, allowing it to self-level, walk, and traverse rough terrain. A novel gait generation algorithm has been created to allow the robot to walk using both static and dynamic gaits. The robot uses a specialized inverse kinematics algorithm in order to translate said gaits into motor rotations. As well, with the torque feedback from the motors, the robot would be able to detect when its feet have touched the ground. With this information, the robot’s control system would be able to adjust the gait so that the robot remains balanced and stable when traversing rough terrain.
As of now, several improvements have been planned, though have not yet been implemented. One such improvement is implementing a vision system in order to assess the terrain ahead of it, and change the gait accordingly, as the current robot is effectively blind. Another improvement is an object avoidance system using AI, possibly with cameras or with LIDAR.