ee106a

Introduction

Design

Implementation: Hardware

Implementation: Software

Results & Video

Reflection

Team

Additional Materials

A Ball Catching 7 DOF Robot Arm

Results

Our project performed reasonably well with our pure ball-tracking implementation. We were able to achieve Kalman filter convergence at around a 20% rate depending on where the ball was thrown (on-shoulder throws had a higher success rate than off-shoulder throws, for example). In these cases, we were able to achieve real-time performance of our main objective. The kalman filter worked very well, converging to an accurate trajectory within ~4 or 5 samples (1/6th of a second).
However, unless ball detection at high frequencies was made sharper by replacing hardware, adding more feature detection capabilities, or resolving lighting in the room, our Kalman filter would be unable to achieve the desired rate of convergence. Currently, we can either achieve real-time performance that sacrifices accuracy or achieve accuracy at the expense of real-time performance. Our Robot controller also was consistently able to move to a given position in real time, and catches were completed when tracking the current position of the ball.

Video Demonstration

Click to Play

All Media files: https://drive.google.com/open?id=1-4PZWXBE6yuuKfYes4byfbNXO1c-ZU1t

Reflection