Please refer to the Path Planning report for detailed analysis.

The report is structured as; section 2 discusses the ACO algorithm briefly and the MATLAB results obtained to support the theory. In section 3 a brief introduction of Artifcial Neural Network and later Backpropagation algorithm is given. In section 4 we have discussed the dynamic vehicle model we have built using mathematical models for path planning controller and in section 5 we have discussed our model and the results obtained. Finally we conclude in section 6 and propose the future expectations.

Robot navigation is governed and managed by a neural network working on the data obtained. We have learned about the neural networks and used back propagation algorithm to train the network. Number of layers in the network are 4. The output of our neural network gives the heading angle to the bot. The trained neural network was successfully implemented and the bot was able to trace the desired path, avoiding obstacles, to reach the final destination. Due to non-ideal setup of the prototype a small amount of noise was observed in the data obtained from the sensors. As a result of which the bot movement was not as accurate as expected at a few instances. In an attempt to improve the accuracy of the results obtained from the noisy data we aim to implement Q-learning or Recurrent learning after learning their application feasibility on our model.

The project was done as a part of Design Project at BITS Pilani under the guidence and supervision of Dr Surekha Bhanot.

Authors : Rasal Kumar, Arjun Gupta, Abhinav Modi