Controlling a Robot Using a Smartphone

Luka Stopar (2013) Controlling a Robot Using a Smartphone. EngD thesis.

Preview

PDF Download (4Mb)

Abstract

In this work, we implemented a motion capture and replication system for human arm movement. Using sensors available on a smartphone, we implemented motion capture and replicated the motion using a NAO humanoid robot developed by the French robotics company Aldebaran Robotics. The motion capture was implemented using the Java Application Programming Interface (API) provided by the Android system. The sensors include a gyroscope, an accelerometer, a magnetometer and finally a video camera. Conceptually, we divided the motion capture into two parts: determining the hand's orientation and determining its location. For the implementation we developed a specialized API containing methods for linear algebra, quaternions and digital signal processing. To determine the orientation of the hand, we implemented two approaches: The first approach uses gyroscope readings to compute the rotation axis and angle with which it updates the orientation. The second approach uses accelerometer and magnetometer readings to construct a reference frame. It then solves a system of equations to obtain the current orientation. We combined the two methods to eliminate drift caused by the first approach and ensure responsiveness. To determine the hand's location we also tried two approaches. The first approach is based on double integration of the accelerometer readings to obtain the location. This approach proved to be unsuccessful, leading to the second approach using computer vision to determine the current location. It tracks four visual features of a pre-defined pattern placed before the device. By knowing the pattern's dimensions, we are able to compute the location. For motion replication, we used the method of differential kinematics. The method computes a pseudo-inverse of the Jacobian matrix with which it updates the current joint configuration. To make the method robust to kinematic singularities we computed a damped least-squares pseudo-inverse of the Jacobian, which uses singular values to define singularity regions in the joint space. It then uses a damping factor in these regions which is proportional to the proximity of the kinematic singularity.

Actions (login required)