Based on the current completion of the project we came up with few ideas that we could implement in future which would lead to better system overall in terms of performance.
1. Eliminate Overhead Localization– The project’s navigation subsystem was solely dependent on overhead cameras. This requires initial setup for the event which is not very optimal. One could do this by using camera to perform visual odometry and fuse that with wheel odometry and IMU readings to get approximate location.
2. Remove April Tag from Human Detection – There are many ways to detect humans some of which are machine learning based algorithms such as Histogram of Gradients method.
3. Make robots appealing aesthetically – Since the robots are intended to be used in social occasions with people around, they should appear more friendly and approachable in nature.
4. Use camera features such as auto-focus and Zoom– One thing we noticed that our output was based on smiles but sometimes picture of the person would be blurred and robot has to come closer to the person to capture his expressions. One could simply zoom instead of going closer to the person and also employ auto- focus for better quality.