Research Group
Sarah Dean's Research Group
- Role: Undergraduate Researcher
- Duration: June 2024 - Present
- Description:
- Rewrote a C++ balloon simulation library, an engine written by Windborne System's balloon simulation engine, in Python JAX for balloon navigation simulation
- Translated auto-differentiation support previously provided by Enzyme into JAX with similar speed
- Writing a Python library called atmosnav, fully supported with JAX, for reinforcement learning (RL) and optimal control in atmospheric agents simulation
- Investigate applying Model Predictive Control to atmospheric balloon control versus applying deep reinforcement learning. Benchmark other data-driven methods with Google's Balloon Learning Environment(BLE).
- Investigated and simulated the nonlinear control of the affined system using Neural Lyapunov-Barrier Functions in the fully-actuated simple point system and underactuated simple point wind system
- Minimizing tracking error with approximate balloon dynamics for high-altitude balloon station-keeping problem
- Designing and optimizing the cost function for MPC in balloon station-keeping problems under altitude, envelope, and power safety constraints
Research Group
Verifiable Robotics Research Group
- Role: Undergraduate Research Assistant & Undergraduate Researcher
- Duration: Aug 2023 - May 2024
- Description:
- Wrote a Python motion primitive library in ROS that supports multi-threading properties in multi-tasking of each robot and contributed to Hello Robot Stretch open-source codes with multi-Stretch setup
- Implemented the robot control synthesis with discrete logic LTL^ψ in warehouse heterogeneous multi-robot collaboration examples on Hello Robot Stretch RE1 and RE2 robots and Kinova 7DOF Gen3 on Boxer in the project Continuous Execution of High-level Collaborative Tasks for Heterogeneous Robot Teams
- Implemented leader-worker architecture in multi-robot communication across robots produced by different companies
- Set up the user guide of Clearpath Boxer with Kinova Gen3 Arm for the entire lab on Confluence, including both hardware guidance and ROS library
- Trained a multi-object detection model and binary-classification model with YOLO5
Research Group
EmPRISE Lab
- Role: Undergraduate Research Assistant
- Duration: June 2023 - Mar 2024
- Description:
- Designed and built a toy environment with 80/20 for Whole-Arm Manipulation Data Collection and Algorithm Testbed, mounted with Arduino Uno and Linear Actuators
- Compared convergence rate and cumulative rewards of different reinforcement learning strategies, including LinUCB, True Online Temporal Difference λ Learning, and True Online Sarsa λ algorithm in the project Preference-Informed Whole-arm Manipulation for Physical Human-Robot Interaction
- Wrote speech-to-context pipeline for user study, where wakewords trigger the pipeline to listen to the user feedback, transcribed to text, and finally processed by GPT3.5-turbo API as a part of contextual input to the learning agent
- Implemented Torque Control on RCareWorld, a Unity-based HRI simulator developed by EmPRISE
Project Team
Cornell Cup Robotics
- Role: Minibot | ECE Subteam Member
- Duration: Aug 2021- May 2022
- Description:
- Implemented RFID sensor detection and data processing algorithm
- Integrated sensors on the Minibot, including Ultrasonic sensor, RFID sensor, and LiDAR sensor