Here at Millburn Robotics, we've always believed in pushing the limits of VEX. This year, we've taken that a step further. One year of work. 2 bots. 0 drivers.
I oversee the design and programming of our robots, keeping us on track with achievable timelines. I maintain a constantly updating list of clear objectives so that everyone has a job when possible and is aware of what needs to be done and how they can contribute.
I'm the co-captain of AI and helped to manage people and projects as we work our way to creating a fully autonomous robot. I specialize in working on the python side of code, in addition to implementing position mapping.
I worked with the build team on various aspects of both robots, including iterations of lift designs, building of the X-drive and more. I also occasionally worked with the code team, and wrote the code for our PD loops.
I worked on the build team to build different parts of both our 15 and 24-inch bots. I built the frame for our 24-inch bot and its conveyor. I worked on fixing up the wiring for the 24-inch as well. On the 15 Inch, I was responsible for mounting both clamps and spacing out some of the internals. Lastly, I helped with regular maintenance on both bots when necessary.
I helped develop Python code on the Jetson Nano that detects different game elements and sends information to the Vex V5 brain. I also aided in creating a system on the V5 brain that receives data from the Nano and processes it to move accordingly.
I was a part of the code team, working mainly on the YOLO V5 model to detect game objects. I also helped build the 24" bot. In the lead up to VEX Worlds, I created this website as a way to showcase our work.
I help design and implement various small projects in both the building of the robots as well as the code that they run. Some projects include a system to push rings out of the way of our mobile goal lift as well as code that blocks the bots from pursuing rings or mobile goals in inaccessible areas of the field.
I helped develop the YOLOv5 model and integrate it into our system through serial communication. I also helped with coding the robot to make decisions on the validity of detections from the model.
I co-developed an algorithm that would allow the robots to map all the objects they see from the camera using only its position. The robot gets this position using the deadwheels I helped install to have a thorough understanding of where everything is in the field.
I assisted with the build team and contributed to various parts of each robot including the lift, body and more. Millburn Robotics has taught me many things about engineering like the planning with software, trial & error and has made me many more friends.
Worked as a part of the build team, helped design and implement various aspects of the robots, such as the conveyor belt on the 24" bot.
I helped design and build different parts of the robot including the conveyer.
Using the Jetson Nano, we run the YOLOv5 object detection model to generate bounding boxes around mogos and rings from video feed provided by our IntelRealsense LIDAR cameras. These bounding boxes as well as depth readings from the LIDAR give us exact positions of all game elements in frame.
Our program ran at 1.6 FPS November of 2021. With the following speed ups, our program runs at 8 FPS as of May 2022.
A list of all object locations computed by the Nano is sent to the V5 brain through a serial cable connecting the two.
Once a list of data has been sent over to the V5 brain, we cross-check them against “deadzones” that the robot is not allowed to enter. For example, our robot should not be attempting to intake rings located under the balance. Once we determine which objects from this list are valid targets, we select the closest one and tell our robot to navigate towards it.
For simplicity, we define the field as a collection of 36 boxes. Using our robot's position and all YOLOv5 detections, we update each box with information about the number of rings and mogos inside at any given time.
Allows for the robot to have short term “memory” about the layout of areas of the field it has recently seen.
The 24” is designed primarily to fill alliance mogos with rings. This is because each alliance mogo filled with rings is worth 30 additional points. Each filled mogo is then deposited into a corner on our side of the field, where it can be picked up by the 15” and brought onto the balance, or remain in a relatively safe location for the rest of the match.
We use a holonomic x-drive as its strafing capabilities allow great maneuverability. However due to their angled wheels, X-drives are weak and bulky, so we use laser cut wheel hubs designed in CAD to make the most of our space with two motor mounts each for increased torque.
Since our robot is guided by AI rather than human direction, it is highly likely that it will approach game elements like mogos or rings slightly off-center. To counter this, our bot has built in mechanisms that increase our margins of error.
Our mogo possession mechanism uses a 3D printed aligner perfectly shaped around the profile of a mogo to allow for the mogo to “lock into place” as we run into it.
Our ring intake is wider than most other VEX VRC robots to increase our chances of successfully intaking rings.
Unlike alliance mogos, neutral mogos can give points to either team, making them especially important. The 15” is designed primarily to control neutral mogos, using the 15 second isolation period to grab two mogos with its front and back clamps. During the interactive period, it attempts to balance with both mogos on our alliance platform for additional points.
In order to meet the 15” design constraints for this robot, we use custom parts extensively.
