team photo

Figure 1
project photo


Team 2107

Team Members

Faculty Advisor

Paul Simmerling
Paulo Silva
Brendan Sayers

Shalabh Gupta

sponsored by
sponsor logo

Path Planning with Deep Neural Networks

The goal of this project is to build a small-scale car setup employing a variety of different sensors like (depth/camera) to generate ground truth labels that will be used as training datasets for a vehicle maneuvering on a road. We will test and develop machine learning (specifically deep neural networks) that uses the sensor information to construct a vehicle motion in response to the sensor information.

 

To accomplish this goal, a testbed was built that uses Robot Operating System (ROS) to control a sensor-suite capable of self-driving. This suite includes a camera, LIDAR, inertial measurement unit (IMU), and encoders. These sensors are connected to an Arduino/Raspberry Pi that is wifi connected to an external laptop which processes the neural network. This allows for two modes of operating: the training teleoperation mode, and the neural networking mode. In the training mode the robot is controlled by a standard gamepad controller and the data recorded by the car is recorded on the laptop. In the neural network mode, the vehicle drives around the track autonomously.