Haotian Gu's Projects
LeGO-LOAM: Lightweight and Ground-Optimized Lidar Odometry and Mapping on Variable Terrain
Lidar odometry is an effective navigation technique for robot localization when a robot does not have access to GPS. A robot's motion is estimated by solving for the translation and rotation across consecutive lidar scans. For the process to be real-time viable on an embedded system, salient features must be carefully selected from the lidar data, and used as the basis for scan matching. This project will aim to augment a successful lidar odometry framework, LeGO-LOAM, for use with a high-resolution, 64-beam lidar, over which existing algorithms currently do not work in real-time.
Due to its similarities to well-studied image classification and retrieval problems, loop closure has the most potential to be solved with DL techniques. It's also an important issue, as correct loop closures guarantee the consistency of the SLAM map and improve all-around accuracy. Computational efficiency and robustness to false positives are the most important characteristics of a successful loop closure subsystem.
This projects addresses unmanned aerial vehicle (UAV) navigation and path planning under engine-out case for landing under severe weather using a Multi-Level Adaptation approach. This is a milestone in a novel autopilot framework, which enables a UAV under large uncertainties to perform safety maneuvers that are traditionally reserved for human pilots with sufficient experience. In addition, we present a high-fidelity simulation environment for fixed-wing aircraft to test and validate the approach under various uncertainties. For the proposed Multi-Level Adaptation autopilot framework, we present the design and analysis as well as the simulation results for the case of emergency landing due to engine failure under severe weather conditions, a challenging task for an autonomous aircraft.
Sample code for connecting to and configuring the OS1, reading and visualizing data, and interfacing with ROS
Code for the paper "PixelCNN++: A PixelCNN Implementation with Discretized Logistic Mixture Likelihood and Other Modifications"
A map representation based on 3D segments
With simulated landmark and robot, This repository is for implementing 2_D EKF_SLAM. Using observed landmark to assist to estimate the location of robot.