Robocart

01 Background

The Major Qualifying Project, or MQP, is a project at WPI in a student's area of study meant to give the student a real-world engineering experience. In my senior year, I chose to build an autonomous ground vehicle for my MQP with Professor Alexander Wyglinski.

With the price of sensors and computing steadily decreasing year after year, affordable self-driving automobiles are starting to become a reality. This MQP aimed to imagine a divergent take on autonomous vehicle technology by challenging modern vision algorithms rather than pursue expensive sensors like LIDAR. Although sensors like LIDAR have dominated recent ventures, their price, bulk, and vulnerable location atop autonomous vehicles could hinder their uptake in the future. As a result, this project sought instead to pursue vision-based object-recognition and navigation. For this project, a 1995 electric golf cart was acquired for prototyping.

02 Approach

Significant upgrades had to be made to its design in order for it to become autonomous. Because of the large scale of this project, the project was broken down into separate subsystems, each of which were worked on separately to be combined sometime in the future. The different systems requiring modifications are shown below. My work on the project involved the three electromechanical subsystems outlined in orange—the stereoscopic cameras, the automated steering mechanism, and the automated braking mechanism. Each of these subsystems posed its own unique challenges and had to be dealt with differently.

03 Raspberry Pi Camera Mount

Raspberry Pi cameras were used for the steroscopic cameras, since they were cheaply available and readily able to be connected to a wireless network through Raspberry Pi computers. A mounting solution was required for the Raspberry Pi cameras and computers. Creating a mount was more difficult than originally anticipated, since the requirements called for, for example, ensuring that the stereoscopic cameras would always point in exactly the same direction. The final design, after going through five iterations, is shown modeled in CAD below. The red plate holding the Raspberry Pi computers and cameras can be easily removed by pinching a spring-locked mechanism and sliding it out.

04 Automated Steering Mechanism

Next, the steering system needed to be modified to allow the computer to control it. This involved attaching a motor to the dashboard of the golf cart (since there was no space under the hood) using a chain and sprocket system. The appropriate torque and speed calculations were done, and a motor was acquired. A multiturn potentiometer was used for position feedback and was attached with its own sprocket. The potentiometer sprocket was chosen to maximize the resolution of the data coming out of the sensor.

05 Automated Braking System

Finally, the brakes on the golf cart were automated to allow the computer to control them. A non-backdrivable van door motor was chosen for its high torque and low speed. For safety, a new mechanism was designed to allow the brake pedal to override the computer control. The design, shown below in CAD, is such that the motor pulls on the red block with a pair of steel ropes. The red block collides with a collar which pulls the hooked rod forward, engaging the brakes. However, the brake pedal can also pull on the hooked rod without fighting against the motor, making the mechanism backdrivable.