Single Track Trailer Control
Developed a model predictive trajectory generator that produces a trajectory to control a car-trailer system to make the trailer track specified points while the car avoids other specified points. We evaluated this using physical experiments where we wanted the trailer to knock soda cans of choice without the car hitting them first - this was a very cool project!
Description
The objective of the project is to design and build an autonomous single-trailer car that can safely and efficiently navigate through an environment with obstacles.
​
To achieve this objective, we designed a control system using a non-linear trajectory optimization framework and Pure Pursuit controller that can accurately track the car-trailer system trajectory accounting for the non-linear dynamics of the system.
​
To test the control system, we placed obstacles like cans in the environment and built a controller that ensures the car avoids the cans while the trailer is controlled to knock-off the specified cans.
​
​
​
About
The problem is formulated as a trajectory optimization problem in which the car-trailer system is controlled to follow a desired trajectory for the trailer. The trajectory is generated offline using an MPC with the objective that the trailer hit some target obstacles while the car avoids all obstacles. Once the optimized trajectory is generated offline, the car-trailer system is controlled to follow the trajectory using a Pure Pursuit controller.
​
Simulation of the MPC generating a trajectory tracking a given trailer path With the constraints and model approximation we can see that amidst obstacles there is slight divergence from desired trajectory but it is still close enough to knock the physical cans off (except the middle can where the trailer should not be going as per the desired trailer path)
Physical Execution of above simulated path One can see how the car swivels at the first can to generate a whiplash to reach the can with the trailer! This was interesting as we could not find a way manually to be able to knock off the first can without the car hitting it first - but the controller came up with something cool!
Report
The problem is formulated as a trajectory optimization problem in which the car-trailer system is controlled to follow a desired trajectory for the trailer. The trajectory is generated offline using an MPC with the objective that the trailer hit some target obstacles while the car avoids all obstacles. Once the optimized trajectory is generated offline, the car-trailer system is controlled to follow the trajectory using a Pure Pursuit controller.
​
The Team
 |  |  |
|---|