26005 – Cornhole Robot

追加: 2026-05-11

[00:00:03]Hey guys, my name is Samuel Cowen. I'm an engineering management major and the team lead for this project. And I'm excited to introduce you to our cornhole robot. The cornhole robot or CHR was designed to be a fully autonomous system that can compete against a human opponent in a regulation game of cornhole. While other cornhole playing robots exist, the CHR stands out as one of the few capable of performing under professional cornhole rules while costing much less than other models.

[00:00:29]Most notably, the CHR must be able to identify a regulationsized cornhole board and align itself to face the hole, launch a one-pound bean bag to consistently land on the target from 27 feet away, weigh less than 100 pounds, and fit through a standard door frame, bring no harm to the user or spectators when operated properly, and be transportable and operatable by a single user. The CHR is composed of four symbiotic assemblies. A loading assembly, a mechanical frame and rotation assembly, a computer assembly, and a launching assembly. In this video, our team will break down each sub assembly in detail, discuss the software behind the project, and go over our project budget and rollout plan for our project post day.

[00:01:12]>> Thanks, Sam. I'm Clay Perville. I'm a mechanical engineer and the integration team lead. I'll be talking about the launching mechanism. The launching mechanism is the main system of our robot. It consists of a rubber belt, an electric motor, and multiple pulleys to keep the belt in tension as it launches the bag. Our original idea for the launching mechanism consisted of two belts that would sandwich the bag and accelerate simultaneously. This idea was scrapped due to being expensive, complex, and heavy. Our next design only had one belt, but the bag would be placed directly on the top side of the belt and quickly accelerated to launch.

[00:01:39]This idea was built and tested, and we were only able to launch the bag six feet, not even close to our required minimum of 27 ft. This led to us performing a fault tree analysis and identifying multiple points of failure that we can improve upon such as electromagnetic interference, the bag slipping on the belt, and the motor not producing enough torque. We redesigned and rebuilt the system. So now the belt is continuously moving at the required RPM, and the bag is pushed into the launching chamber and launched. This new design blew our distance requirement out of the water, achieving an average of 32 feet at only 80% of our max RPM.

[00:02:10]>> Thanks, Clay. My name is Joel Luna. I'm a mechanical engineer and I'm going to go into the loading mechanism for the robot. We use a top loading system where the bag is placed in a secure area that will be enclosed by a protective panel and safety door to prevent access to moving parts. The bag sits in front of a pushing wall driven by a motor powered rack and pinion system which converts rotational motion into linear motion.

[00:02:31]When activated, the wall moves forward pushing the bag under the conveyor belt.

[00:02:35]The belt does not spin until the bag is being pushed forward, ensuring proper timing and controlled operation. The system uses two limit switches to control motion. One at the forward position to signal the wall to stop and reverse and one at the bottom position to stop it once it returns to its starting position. A kill switch will be installed on the door to stop the system if the door is open during operation.

[00:02:55]Overall, this mechanism provides a simple, safe, and controlled way to prepare each bag for launch.

[00:03:03]Thanks, Joel. My name is Jack Krill, and I'm a mechanical engineer major. One major aspect of the functioning robot is the rotational mechanism. This mechanism is the connection between the base frame and the launching mechanism that rotates to aim the launch towards the board. The main component that allows the connection and the rotation is the thrust bearing that supports the axial force from the launching mechanism while allowing it to rotate. This bearing is then working with the two 3D printed enclosures we designed which serve the purpose of one being a static on the bottom attached to the base frame while the other both attaches to the launching mechanism and interacts with the gear which turns it. The gear is also a 3D printed piece which we designed which attaches directly onto our stepper motor and meshes with the teeth on the thrust bearing enclosure we designed. An important feature we added later in the process as we found a need for it was support rollers. Once we had assembled and connected the launching mechanism to the base frame, we realized that there was a lot of room for the top half to vibrate and recoil when it would launch back. Our solution to this was to add a supported surface on the base frame and then round rollers onto the launching mechanism, which would, like the thrust bearing, support the downward force from the launching mechanism while still allowing it to rotate.

[00:04:13]>> Thanks, Jack. I'm Colin Brown, the team's optical engineer and documentation lead, and I designed and integrated the systems electronics. For our sensing system, I chose a MOCO's 4K USB webcam with an IMX 678 sensor and a telephoto verifocal lens. This feeds into our Raspberry Pi 5. Later, Andy will get into our vision approach. The main launch motor of the Cornhole robot is an Orive M8325S driven by an Orive Pro. This combination is capable of 100 amp bursts or almost 4.5 kW with our system voltage. Powering the robot are two 6S lipo batteries safely contained in a fireproof lipo bag. These batteries connect to our power distribution box which puts them in series and provides our system with a nominal 44.4 volts.

[00:04:55]Inside the enclosure is a USB power supply and voltage readout, 80 amp fuse and a contactor controlled by our estop capable of breaking the circuit under full load. The main voltage output goes directly into the O drive pro as well as our stepper motor driver for the rotation mechanism which is driven by a Nema 23 stepper motor. The loading mechanism is powered by a simple 24volt DC motor and Hbridge motor driver requiring a 24volt buck converter.

[00:05:18]Controlling all of our electronics is the Raspberry Pi 5. The Pi is contained in a custom enclosure with a 7-in touchscreen providing convenient control for the operator. The O drive is communicated with over canvas, a reliable and interference resilient automotive standard. The loading mechanism, stepper motor, and limit switches are all controlled with GPI opens.

[00:05:37]>> Thanks, Colen. My name is Andy. I'm a self-engineer major.

[00:05:41]>> Our system runs entirely on device using machine learning. No cloud, no APIs or no data ever leaves the device. It's powered by custom training YOLO VA network built on 2,000 plus handle label images each annotated with precise key points for the four board corners and whole center. The model was trained across diverse real world conditions including varieting lightings, distances, camera angles, shadows and board colors to ensure a strong generalization.

[00:06:06]The model perform real-time object detection and post estimations directly on Raspberry Pi. To achieve high precisions, we implement a twostage hybrid approach. The neuronet networks provides a robust detection while a CIE76 colorbased refinement layer extract the board boundaries with sub pixel accuracy. In short, machine learning at the targets and classical computer vision refineses it resulting in consistent reliable performance without manual tunings. The system is operated through a custom KV dashboard on a 7in Raspberry Pi touchscreen designed for fast and safe interaction with large controls such as start, estop, and kill. It also includes onscreen keyboards, a live terminal, and one tap switching between manual and autonomous mode. In manual mode, each subsystem, flywheel, loader, stepper, and camera can be controlled independently for testing and calibrations. In autonomous mode, a single tap executes for the float pipelines, self-centering, detection, trajectory calculations, angle adjustments, aiming, and firing. After initial shots, both RPM and aiming can be adjusted in real time to account for environmental factors such as wind with both measured and ply values displayed for full transparencies.

[00:07:13]>> No laptop, no knowledge, just come to robots, tap, and run.

[00:07:19]>> Thanks, Andy. I'm Josh now. I'm the engineering management major and I'm also the procurement lead. So for our project, we were able to stay within our 4,400 budget. Our final spend came out to about 3,700, leaving us with roughly $700 remaining. Breaking that down, about 61% of our budget went towards mechanical components, 33% toward electrical, and about 6% toward optics.

[00:07:40]To stay in our budget, we used a combination of purchase parts and fabricative components. We bought key items like motors and electronics while also utilizing customized 3D parts. This allowed us to iterate quickly and avoid unnecessary costs from constantly reordering parts. For our delivery plan, after design day, we will transport the Cornhole robot to the Rathon bike shop and deliver it directly to our customer.

[00:08:01]We'll provide a user manual that covers operation, safety, and basic troubleshooting. And we'll also walk them through a quick demo to make sure they're comfortable using the system.

[00:08:09]Thank you guys for watching. Can't wait to see you guys at Design Day. Go Cats.