Software

Analysis
Following a careful examination of the problem it was decided to divide the software design into the following three areas:

• Movement: A set of functions that would control the robot’s movement around the arena. These would include simple functions such as those to make the robot move forwards, as well as higher-level functions such as those to prevent the robot from crashing into walls, or those to make it go to a specific corner. They would be triggered by events occurring inside the arena, such as the successful collection of an object.
• Control: This set of functions would be developed to control movements in the robot that do not impact on its physical location. As an example, functions would be required to open and close the robot’s arms, and also to control the IR sweep.
• Object Detection: The final group of functions would be generated for the purpose of object detection. They would be required to detect the presence of an object, its location relative to the robot, and also to provide a positive identification of the object following its collection.

Movement
The initial analysis carried out prior to the final software design considered the most appropriate patterns of behaviour for our robot. One area that was investigated in detail using experimental methods was the best way to implement a turn function for the robot. Two alternative approaches were proposed:

1. Upon encountering an object or wall the robot should make small turns incrementally until it returns to a state that is
....acceptable
2. Upon encountering an object or wall, determine the incident angle and turn an appropriate amount immediately.

For patterns of movement it was decided that when searching for an object an approach should be taken that copes well with the random distribution of the objects. For this reason, it was decided that the most suitable approach would be to create a simple function that turned the robot in a random direction when it hit a wall. In contrast, when the robot is carrying an object, the wall should be followed in order to help the robot find the appropriate corner.

Corner Detection
For corner detection it was decided that the simplest solution would be to take a reading of the colour of the floor when the robot hit a wall. If the reading indicated that the robot was in a corner, then it could act accordingly, otherwise it would follow the wall until it did reach a corner.

Object Collection vs. Wall Avoidance
A design problem that became apparent immediately upon embarking on the project was the requirement for the robot to clearly distinguish between objects in the arena and the walls. Since the walls of the arena were much taller than the objects contained within it an obvious solution seemed to be to use two IR sensors placed at different heights on the robot. The lower IR sensor would be used to detect objects within the arena, the higher one to detect the walls. However, two sensors presented the additional problem of deciding which sensor should have precedence in influencing the robot’s behaviour. An approach was taken to allocate priorities to the respective sensors, with the higher sensor given the highest priority. This was decided since the higher sensor would only ever detect walls and never objects. Since avoiding walls (which could cause the robot to get stuck) was a higher priority, triggering the higher sensor would result in the robot acting to avoid a wall, satisfying the requirement of not getting trapped on a wall or in a corner.

Feasibility

Movement
Following some basic experimentation it became clear that the first approach involving incremental turning would result in the robot moving around the arena unacceptably slowly. Therefore, it was decided that the second approach would be the most suitable. The relationship between the turning speed of the robot and the angle to be turned was modelled. A function was then written that would convert the incident angle, as approximated by the IR sweep, into a time in milliseconds which the robot should turn for. This worked consistently well, and was incorporated into the final design.

Experimentation on the idea of turning the robot in a random direction when it hit a wall when searching for objects proved feasible. The robot demonstrated good coverage of the arena, indicating that this approach would lead to a high success rate in detecting objects.

Corner detection
After initial feasibility testing, it as discovered that problems differentiating between the red corner and the floor meant that this method of corner detection would not perform sufficiently well on its own.
In an attempt to find an alternative approach the sweep values for the IR sensors were analysed when the robot was placed in a corner and compared to sets of values obtained when the robot was next to only one wall. It was found that a sweep conducted when the robot was placed in a corner resulted in an array containing two maxima. This provided a secondary condition from which we would be able to determine if the robot was in a corner.

The values obtained from the sweep did not always rise and fall smoothly however. Anomalous values seemed to occur relatively regularly so a decision was taken to perform smoothing on the array obtained from the IR sweep. To perform the smoothing, each value in the array was replaced with an average of itself and the two values either side of it. As a result of the smoothing it appeared that the robot began to identify the presence and location of walls more accurately; it also improved its ability to distinguish between a single wall and a corner.
As part of the smoothing the lowest and highest indexes of the sweep array were reset to zero. This was due to the unusually high incidence of high values detected when no object was present at these locations. This often caused the robot to decide that it was in a corner when in fact it was close to only one wall. Since the highest and lowest indexes of the array are gathered from a direction approximately perpendicular to the robot’s direction of travel, no negative behaviours were observed as a result of this practice.

Object Detection vs. Wall Avoidance

The idea of using two IR sensors yielded promising results following some initial tests in the robotics labs. Initially, some problems were encountered due to the higher sensor being triggered by objects placed in the arena. When this situation occurred the robot would avoid the object instead of turning towards it as desired. The problem was solved by raising the highest IR sensor slightly and by ensuring it was placed so that it scanned horizontally. Following this adjustment, the observed behaviour of the robot improved and it seemed to quite accurately distinguish between the walls and the objects that should be collected. This means of using two Infra Red Sensors was therefore incorporated into the final robot design.