Tennis ball robot

Design Team Members: Eric Hunsberger, Steven Leigh, Diego L. Pai, Michael Smart

Supervisors: Dr. David A. Clausi

Background

Tennis is fun and provides good exercise. However, repetitive ball collection can prove to be quite annoying. In this regard, our 4th year systems design engineering design team has decided to take on the challenge of finding an adequate solution to the issue of tennis ball retrieval. Some ball collection devices already exist in the market and can help players reduce the time spent picking up balls. However, these devices are manually intensive and require the direct involvement of the users. If ball retrieval is automated, players will spend absolutely no time collecting balls! Therefore, the issue our team is concerned with is the automation of ball retrieval: To reduce the time taken to collect tennis balls, we want to design and construct a robot that localizes balls, collects them and then returns them to the player.

Features

To successfully design and build the tennis ball robot, our design team needs to solve many pieces to the puzzle. The specific robot design requirements are:

  • Suitable frame and drive
  • Ball detection device
  • Ball collection mechanism
  • Ball storage capacity

In addition, an impressive set of intelligent systems need to be implemented for the design’s capability to be autonomous:

  • Localization intelligence
  • Scanning and searching
  • Path planning algorithm
  • Path following algorithm

Design methodology

To choose the most appropriate collection mechanism, several design alternatives were compared based on specific criteria such as size, power requirements, ease of implementation, collection speed and cost. The chosen alternative is the Modified Conveyor Belt collection mechanism. Figure 1 illustrates how this mechanism works:

Tennis ball conveyer belt collection mechanism

Figure 1: Modified Conveyor Belt collection mechanism

Furthermore, ball detection is essential to the project as it allows the robot to estimate its heading and distance vis-à-vis a detected ball. The method applied to detect tennis balls is called the circular Hough Transform. This is a common method used to detect general round objects. Figure 2 below illustrates a particular application of the circular Hough Transform:

Circular Hough transform

Figure 2: Applied circular Hough Transform

Filtered circular Hough transform

Figure 3: Filtered circular Hough Transform

Several improvements can be made to the circular Hough Transform in order to obtain a superior ball detection algorithm. For instance, by changing the image gradient, the detection rate can be significantly improved. Also, by observing that the tennis balls can only be found on the physical ground (as opposed to in mid-air), the detection speed can be greatly increased. Another improvement to the detection algorithm is to add a filter that ensures all detected objects are in fact round. Figure 3 above shows a rendering of the filtered circular Hough Transform method. As you can see, the filter managed to remove the majority of the false positives (detected objects that are not actually round).

These aspects are just a fraction of the interesting design considerations our design team needs to take into account to fully design and build the tennis ball robot. Other interesting aspects involve vehicle hardware design and conception, intelligent systems programming and implementation and much more. Once all the design challenges are overcome and all the pieces to the puzzle are solved, our tennis ball robot will be collecting balls as users play tennis, thus making the game more enjoyable!