roboticProject

Nowadays, online learning is not yet widely implemented for Human-Robot Interaction due to a few difficulties imposed on algorithms. One of the issues is the random exploration technique required for efficient exploration of state space that results in unpredictable non-collaborative behavior. A slow convergence to the optimal solution is yet another issue. To overcome these two deficiencies, we introduce cheap-talk communication between user and robot that helps the online-learning algorithm to achieve effective collaboration between the two players. Experiments with humans demonstrated that  feedback and planning cheap talk can substantially improve a robot’s ability to learn to interact with a human partner in situations when both partners' goals do not totally coincide.

I was responsible for developing a computer vision system for a robot playing the block-sharing game. In these game settings two players share a set of nine blocks and take turns to pick up a block until they each have three blocks, with one player going first in each round. If a player’s three blocks form a valid set (i.e., she has all blocks of the same color, all blocks of the same shape, or none of her blocks have the same color or shape), then her payoff in the round is the sum of the numbers on her blocks. If she fails to collect a valid set, she loses the sum of her blocks divided by 4.

The problem was to teach the robot how to differentiate between different shapes (triangle, circle, square) and colors (red, blue, yellow). The system I have devised with the help of OpenCV Computer Vision library proved itself to be quite robust and performed well even in different lighting conditions.

The gameboard observed by the robot before starting shape recognition.

Almost all the shapes are recognized.