Tianyi Gu's Home Page

Experience

University of New Hampshire RA, TA | 6 years

I have been a research assistant at the UNH AI Group. I conduct theoretical and empirical research projects to design principled algorithms for autonomous agents to plan in an uncertain environment under time pressure. My works have been published in several AI conferences such as AAAI, IJCAI, ICAPS, and IROS. Please see the publication session below for more details.

I have also been a teaching assistant at the University of New Hampshire for a variety of courses, involved in creating assignments and exams and conducting recitation sessions for Algorithms (C), Intro to AI, Intro to Computer Science (Java, Python), Intro to Software Engineering (DevOps tools), Intro to Computer Security, Database Programming (C#, SQL), Scripting Languages (bash, zsh). Here are some teaching samples.

Motional Summer Research Intern | 12 weeks

I was a research intern in the planning team at Motional (Aptiv's self-driving team) in the summer of 2020. I proposed and implemented a learning-based approach to enhance the planner. The feature was integrated into the next-generation planner. According to the non-disclosure agreement, I can talk more details only after our in-progress patent is granted.

Cognitive Assistive Robotics Lab Summer Research Intern | 12 weeks

I was a robotics intern at CARL at UNH in the summer of 2019. I worked on a proof-of-concept research project that builds a socially assistive robot to support the caregiving of individuals with Alzheimer's disease. In the project, I

Build a smart-home-based service robot framework that can provide real-time Alzheimer's disease care.
Build an AI planner based on ROSPlan that performs real-time online task planning.

Two papers are published from this project: a robotics paper and a gerontology paper. Here is my talk at ICAPS and slides. Here is a video that shows two caregiving scenarios: 1) the robot reminds the patient to take medication, 2) the robot is preventing a dangerous walkaway of the patient. The second part of the video record a group of real-world caregivers participating in our user study. I also demoed the system to a group of undergrads. All of the source codes are here.

Realtime Robotics Summer Research Intern | 12 weeks

I was a robotics intern at Realtime Robotics in the summer of 2018. I worked on a motion planning project that could enable an autonomous vehicle to safely drive in crowded urban areas and also achieve the goal regions as quickly as possible. In the project, I

Build a real-time planner for an autonomous vehicle that can safely drive in a crowded urban area. The planner was lattice-based and performed an anytime search.
Build a real-time planning framework that enables handling a dynamic world online.
Build a simulation environment to demonstrate flagship products to significant new customers.

Here is a video shows the vehicle avoids hitting a man who rushed into the road. Here is another video that shows that the car obeys the traffic light. It also visualizes the online lattice is also. Here is my talk about this project and slides.

Port of Shanghai Software Engineer | 3 years

I worked as an operations research engineer and a software engineer at the Shanghai International Port Group (SIPG, 上港集团) for three years (2012-2015) after I graduated from the Shanghai Maritime University with my Master's Degree in Logistics Engineering. While I was at the SIPG, I worked on the following projects:

Member of the team that designed, built and deployed a new automated container terminal operations management system, including algorithm development for the crane allocation and scheduling module and the financial module.
Helped launch previous terminal operating management system.

Alcatel-Lucent Software Engineer Intern | 6 months

I worked as a software engineer intern at Alcatel-Lucent (Shanghai)(上海贝尔) during the summer of 2011. While there, I developed a global electrical elements database, including web interfaces and database maintenance software.

Peer-reviewed Publications

Maximilian Fickert*, Tianyi Gu*, and Wheeler Ruml, Bounded-Cost Search Using Estimates of Uncertainty. Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence (IJCAI-21), 2021.
[pdf]
Maximilian Fickert*, Tianyi Gu*, Leonhard Staut*, Sai Lekyang, Wheeler Ruml, Joerg Hoffmann, and Marek Petrik, Real-time Planning as Data-driven Decision-making. Proceedings of the ICAPS Workshop on Bridging the Gap Between AI Planning and Reinforcement Learning (PRL-20), 2020.
[pdf] [publisher] [slides] [poster] [talk] [code]
Tianyi Gu, Momotaz Begum, Naiqian Zhang, Dongpeng Xu, Sajay Arthanat, and Dain P. LaRoche, An Adaptive Software Framework for Dementia-care Robots. Proceedings of the ICAPS Workshop on Planning and Robotics (PlanRob-20), 2020.
[pdf] [publisher] [slides] [video] [talk] [code]
Maximilian Fickert*, Tianyi Gu*, Leonhard Staut*, Wheeler Ruml, Joerg Hoffmann, and Marek Petrik, Beliefs We Can Believe In: Replacing Assumptions with Data in Real-Time Search. Proceedings of the Thirty-fourth AAAI Conference on Artificial Intelligence (AAAI-20), 2020.
[pdf] [publisher] [slides] [poster] [code]
Sajay Arthanat, Momotaz Begum, Tianyi Gu, Dain P. LaRoche, Dongpeng Xu, and Naiqian Zhang, Caregiver Perspectives on A Smart Home-based Socially Assistive Robot for Individuals with Alzheimer's Disease and Related Dementia. Disability and Rehabilitation: Assistive Technology, 2020.
[pdf] [publisher]
Bence Cserna, William J. Doyle, Tianyi Gu, and Wheeler Ruml, Safe Temporal Planning for Urban Driving, Proceedings of the AAAI Workshop on Artificial Intelligence Safety (SafeAI-19), 2019.
[pdf] [publisher] [slides] [poster]
Reazul H. Russel, Tianyi Gu, and Marek Petrik, Robust Exploration with Tight Bayesian Plausibility Sets, Proceedings of the 4th Multidisciplinary Conference on Reinforcement Learning and Decision Making (RLDM), 2019.
[pdf] [poster]
Scott Kiesel, Tianyi Gu, and Wheeler Ruml, An Effort Bias for Sampling-based Motion Planning, Proceedings of the IEEE/RSJ Conference on Intelligent Robots and Systems (IROS), 2017.
[pdf] [publisher] [video] [talk] [slides] [code]
Yi Ding, Xujun Wei, Yang Yang, and Tianyi Gu, Decision Support-based Automatic Container Sequencing System Using Heuristic Rules, Cluster Computing 20(1) 239-252, 2017.
[pdf] [publisher]
Chengji Liang, Miaomiao Li, Bo Lu, Tianyi Gu, Jungbok Jo, and Yi Ding, Dynamic Configuration of QC Allocating Problem Based on Multi-objective Genetic Algorithm, Journal of Intelligent Manufacturing 28(3) 847-855, 2017.
[pdf] [publisher]
Yuping Wang, Yangyang Hao, Yuanhui Zhang, Youfang Huang, and Tianyi Gu, Berth Allocation Optimization with Priority based on Simulated Annealing Algorithm, Journal of Engineering Science & Technology Review 11(1) 74-83, 2017.
[pdf] [publisher]
Yi Ding, Shuai Jia, Tianyi Gu, and Chung-Lun Li, SGICT Builds an Optimization-based System for Daily Berth Planning , Interfaces 46(4) 281-296, 2016.
[pdf] [publisher]
Chengji Liang, Tianyi Gu, Bo Lu, and Yi Ding, Genetic Mechanism-based Coupling Algorithm for Solving Coordinated Scheduling Problems of Yard Systems in Container Terminals, Computers & Industrial Engineering 89 34–42, 2015.
[pdf] [publisher]
Yi Ding, Tianyi Gu, Guolong Lin, and Chengji Liang, The Establishment and Solution of Coupling Model on Coordinated Scheduling of Handling Facilities in Container Terminals, Applied Mathematics & Information Sciences 6(3) 915–924, 2012.
[pdf] [publisher]

Research Visits and Invited Talks

2021 Invited talk at the IBM Thomas J. Watson Research Center.
[talk] [slides]
2018 Guest lecture for the University of New Hampshire's CS900: Graduate Seminar.
[talk] [slides]
2018 Invited lecture at the University of New Hampshire's Robotics Seminar Series.
[talk] [slides]
2018 Invited lecture at Shanghai Maritime University's Logistics Research Center.
[talk] [slides]
2017 Invited lecture at the University of New Hampshire's Robotics Seminar Series.
[talk] [slides]

Coursework Projects

In CS880 Introduction to Mobile Robotics, my final project studies the problem of autonomous mapping. We implement the frontier-based exploration algorithm combined with the occupancy grid mapping technique that enables a Turtlebot robot to build a map autonomously for an unknown environment. We applied the theory of Bayesian inference to update an occupancy map. We also used the frontier based exploration algorithm to navigate the robot to unexplored areas on the map. The experiment results show that the robot can map the environment fully autonomously, both in simulation and in real-world settings. Our source code is here. We also took a video that records the process of a Turtlebot mapping a corner area in Kingsbury.
In CS980 Topics in Reinforcement Learning, my final project studies the problem of dynamic obstacles avoidance for mobile robots. We studied two deterministic approaches that use heuristic search techniques and four stochastic methods that use reinforcement learning techniques. We proved these two types of systems are mathematically different. The experiment results show that deterministic approaches are not only faster but also more robust than stochastic approaches. But stochastic methods often applicable for specific problem scenarios. Our source code is here.
In CS980 Planning for Robots, my final project designs two control algorithms: a sampling-based model-predictive control (SBMPC) and a bisection search-based model-predictive control (BBMPC). We build controllers on top of these algorithms for a real-time planning system that enables a Pioneer robot to move quickly in environments with dynamic obstacles. We use the ROS platform to establish communication between different modules (i.e., perception, planner, executive). We run experiments in both simulation environments and real-world world environments. The investigation shows that the algorithms have pros and cons, respectively, depending on the path’s curvature. We also discussed several issues with the real-time planning system. Our source code is here.
In CS830 Introduction to Artificial Intelligence, my final project presents a new anytime motion planning approach called B-SST. B-SST first runs BEAST, an effort-aided planner, to find a suboptimal solution as quickly as possible. Then it switches to another motion tree growth process called SST-with-cost-pruning that adopts both ideas from SST and cost pruning algorithms. We first introduce several related works. Then B-SST is described in detail. Results with various vehicles and environments showed that B-SST is competitive compared to A-BEAST and other successful anytime planners. We also discussed a more sophisticated idea of creating a better anytime motion planner in the end. Our source code is here.
In CS880 Introduction to Information Retrieval, my final project studies the task of helping an AI player win a computer game by reading a strategic user's guide designed for human players. In complex computer games such as StarCraft, WarCraft, and Civilization, finding a winning strategy is challenging even for humans. Therefore, human players typically rely on manuals and guides. Recently, researchers have tried to use such textual information to train an AI player. Our goal is to better understand the retrieval models used in their paper. Our results provide evidence that shows the basic inverse document frequency (IDF) can, surprisingly, often outperform approaches that rely on recurrent neural networks (RNN). Our source code is here.
In CS880 Probabilistic AI and Machine Learning, my final project compared the accuracy and reliability of several different classifiers for recognizing handwritten digits, training on MNIST dataset. We implement classifiers such as K-Nearest Neighbors, Decision Tree, and Random Forest for comparison. Our results suggest they both have their pros and cons. Our source code is here.
In CS980 Topics in Multi-Agent and Multi-Robot Systems, my final project was an MDP-based approach to solve the Quay Crane Scheduling Problem (QCSP) under uncertain demand. In container terminals, the number of time-conflict tasks for yard cranes (YC) in yard operation is considered one of the critical measures for evaluating the level of fluency for quay crane scheduling by terminal experts. We model QCSP in MDP, design a reward function to minimize the number of time-conflict tasks, and apply the UCT algorithm to solve the model.

Tianyi Gu