Spet 2023-Dec 2023
• In this project, we established a comprehensive Automatic Parking System with known parking lot map
• Construct kinematics bycicle model • Build dynamics model with linearization for car using the Nonlinear Tire Model Magic Formula
• Implemented and tuned linear and non-linear MPC controller with quadratic cost function
• Deply Hybrid A* for path planning with high-dimensional state spaces allows for the generation of smoother and handle realistic motion constraints
• Evaluate performance with deploying simulated user-defined parking lots with car number and layout using CV2

Mar 2023-May 2023
• Compared Panda robot arm planar pushing task behavior on discrete numerical differential equation based networks (ResNet, PolyNet and FractalNet) and continuous neural ordinary differential equation based network.
• Evaluated learned models on an open-source forward dynamics dataset collected by using KUKA LWR and Baxter.

Mar 2023-May 2023
• Reproduction of Loc-NeRF, a real-time Monte Carlo localization method uses NeRF as a map representation.
• Achieved efficient localization performance with adaptive particle filter by using KLD-Sampling method.
• Evaluated on both indoor (LLFF) and novel largescale outdoor (OMMO) NeRF datasets.

Apr 2023-May 2023
• Reproduction of the IEEE ICRA 2019 Best Conference Paper "Making Sense of Vision and Touch: Self-Supervised Learning of Multimodal Representations for Contact-Rich Tasks".
• Make robot be able to perform contact-rich manipulation tasks by using multimodal information.
• Learn a model for multimodal representation.
• Demonstration of insertion tasks from panda simulation enviornment.
• Implement and compare across several algorithms to compute multimodal representation as latent layer: addition, concatenation, adaptive weighted average, and Gaussian predicted.

Sept 2022-Dec 2022
Computer Vision:
• Developed camera (Realsense L515) intrinsic and extrinsic calibration using Kabasch algorithm.
• Automatic workspace reconstruction using affine transformation of workspace grids, and masking out the robot arm during operation.
• Using CIE Lab color space for block color identification.
• Implemented the method of focusing on the topmost block on each stack of blocks.
• Evaluated accuracy of camera calibration using heatmap
Kinematics:
• Implemented Denavit-Hartenberg forward kinematics and geometric inverse kinematics for RX-200.
• Deployed gravity compensation by modeling the defection in each joint as a linear spring
• Designed gripper for potential replacement of the original one on RX-200 with 3D Modeling
Competition:
- 9 blocks, random sizes, random colors, possibly stacked
- Pick 'n stack: Stack up the large and small blocks in separate stack in rainbow color order, scored on completeness and correctness of the order.
- Line 'em up!: Line up the large and small blocks in separate lines in rainbow color order, scored on neatness and correctness of the order.
- To the sky! : Stack up the large blocks in any color order, as hight as possible, scored on number of blocks.
*** Results: Second place amoung 15 teams.

Sept 2022-Dec 2022
Motion & Odometry
• Developed wheel speed characterization and PID controllers for wheel speed and robot frame with odometry and gyro sensor fution
SLAM
• Constructed an Occupancy Grid Mapping using the Bresenham's algorithm along LiDAR rays.
• Implemented Monte Carlo Localization and A* Algorithm for path planning.
• Deployed Map Exploration with algorithm for finding the frontiers and compared between BFS and DFS efficiencies
Competition:
• Looper: From the starting pose, make 4 circuits around a convex arena by driving the path given using localization only. Return to the starting pose perfectly.
• Dungeon Explorer: From the starting position, explore the complex maze and make a map of the environment. Then return to the start location.
***Results: First place in Looper event, second place in overall competition among 15 teams.

Sept 2022-Dec 2022
• Designed a vision-based system improves accuracy of product detection in Autonomous store at low-light conditions.
• Collaborated with three other projects to construct a robust enhanced visual detection system containing: an Unreal Engine simulation for training data, mask extraction of hand pose, and best-camera-angle selection etc.

Mar 2021-April 2021
• Prototyped robotic arm version of Beer Pong party game to achieve almost perfect accuracy.
• Designed a 4-joints robot arm to aim and struck the ping pong ball to target cups with given coordinates using inverse kinematics, projectile motion, and feedback control.