University of California, San Diego
Mechanical and Aerospace Engineering
MAE 156B: Senior Design Project
Background
Recent advances in artificial intelligence (AI) and learning in robotics have opened possibilities for a “key-in-lock” challenge, which represents a step forward in precision, robotic learning, and adaptive control techniques compared to the previous “peg-in-hole” challenge. Towards this end, the team was assigned the objective of developing a test stand capable of gathering data on the positions, orientations, forces, and torques involved in opening a lock with a key, as well as eventually combining AI with Robot Programming by Human Demonstration (RPHD) to solve complex problems and perform difficult tasks.
A testbed including a lock with “easy” and “hard” keys was developed, and it currently supports a 6 degree of freedom (DOF) optical position/orientation sensor and a 6 degree of freedom (DOF) force/torque sensor. These signals are collected over time and plotted after the data acquisition is completed. Additionally, a robotic assembly including a Stewart platform, servo motor, and linear actuator was designed to mimic 6 DOF human motions with a key.
The ultimate goal of replicating the human unlocking motion was achieved by developing and creating the mechanical hardware, sensor package, robotic assembly, and testbed, and creating a master controller that executes the unlocking motion. The product created by the team will enable future AI researchers and scientists to apply the RPHD AI approach to increase the robot’s ability to complete complex human tasks and progress the field of precision robotics.
Project Requirements/Priorities
First Priority:
Build testbed that can precisely record 6 DOF forces/torques and 6 DOF positions/orientations associated with a key unlocking a lock
Include purposeful variations, such as grinding keys, so the task is difficult
Second Priority:
Build a robotic arm that has the DOF to perform the unlocking task, including key wiggles
Reach Goals:
Use the RPHD AI approach to learn and then implement a peg-in-hole task
Use the RPHD AI approach to perform unlocking with known key
Use the RPHD AI approach to perform unlocking with unknown keys
Description of Design Solution
Sensor Package
Optitrack Sensor
Tracks the position and orientation of the infrared (IR) markers in real-time at 120 Hz with sub-millimeter accuracy. This data is essential for understanding the spatial relationships and movements involved in unlocking the lock.
ATI Mini 45
Tracks the amount of force and torque needed to unlock the lock. Helps the robot learn the necessary amount of force to apply when inserting the key into the lock and turning it, ensuring that the robot can replicate the task with appropriate force control.
Robot Components
Linear Rail
The robot uses a ball bearing linear rail system to translate the robot's position, achieving the large insertion/retraction motions of the unlocking sequence.
Stewart Platform
The Stewart Platform enables the robot to position the key with high precision in 3D space and is used to replicate the smaller jiggling motions of the unlocking sequence. It's motion can be programmed using Cartesian coordinates and Euler angles from human-gathered data through the OptiTrack camera sensor.
Dynamixel (Roll Servo)
The roll servo motor mounts on top of the Stewart platform and replicates the large rotational action of the key unlocking sequence.
Narrated Videos
Our Sponsor
Our Team, Marfred Barrera, Kerseyleanne Catolos, Rachel Hartanto, and Johnny Li, with our sponsor Professor Nathan Delson. Dr. Delson is a Professor at UC San Diego Mechanical & Aerospace Engineering (MAE) Department. His research and interests lie in robotics, controls, project-based learning, and AI. Professor Delson seeks to explore the possibilities of AI in precision robotics.