I joined FAU as an assistant professor in August 2023 to continue to pursue my research in robotics. Here is my new personal website.

I started as a software engineer at Google in Mountain View, CA in June 2022. I was part of the motion tracking team, working on improving the localization and tracking performance of AR products.

Previously I worked for iRobot Corporation starting January 2022 as a Senior Robotics Software Engineer. My general role was to develop mapping and planning algorithms for iRobot products to improve their cleaning performance. 

I graduated with a Ph.D. degree in Computer Science from the University of Minnesota, Twin Cities. My advisor is Prof. Volkan Isler. I worked on optimization in robotics under practical constraints, with a primary focus on energy efficiency. Many field robots are powered by batteries and have limited energy budgets. To have them work persistently in fields, we must consider the energy constraint. My research tackles this constraint using both algorithmic and deep learning-based methods. I joined the lab in September 2016. Contact: weixx526@umn.edu

Ph.D. Research

1. Energy-aware coverage path planning

If a robot with energy constraints wants to cover a large area, it may not be able to fully cover the area in one iteration. It needs to visit a recharging station to get recharged and then continue the work. Multiple paths need to be planned, and the robot needs to be recharged between two paths. 

We have presented two approximation algorithms to this problem with theoretical proof for the performance compared to the optimal solution [1]-[2].

2. Air-to-groud cooperation for Energy-efficient path planning of ground robots

Another challenge in the applications of field robots is the disconnect between current theoretical results and practical applications due to the difficulty of obtaining maps for path planning. 

We build the maps for large, non-uniform environments by combining the aerial images with the ground robot energy-consumption measurements. Deep networks are applied to learn from the collected data and predict the energy-cost maps for the entire area [3-5], [7].

3. Occupancy map inpainting for online robot navigation

Indoor navigation using sensors with limited field of view and occlusion is a challenging task, especially for small-sized robot.

We use a learning-based method to predict the occupancy of unseen areas around the robot (occupancy map inpainting) [6]. The training data is collected using a two-camera setup, where the high camera, which can see a larger area, is used to supervise the training. With the inpainted occupancy map, the robot reaches goal locations faster, compared to using the raw maps.

4. A practical robotic system for agricultural weed control using autonomous robots

Pastures:

Cows on pastures usually eat good grass and leave weeds behind. I worked on developing navigation methods for the mower to autonomously mow the pasture (video link).

Using robotic mowers instead of chemical herbicides can not only reduce costs, but protect our natural environment.

More information 

Cornfields:

Cornrows tend to be narrow and it is hard for human to do in-row operations. One of my projects is to develop navigation methods for the robot (in the picture) to travel through corn rows to collect data. Weed control can be performed with an onboard camera to detect weeds.

Publications

1. M. Wei, V. Isler, Coverage Path Planning under the Energy Constraint, International Conference on Robotics and Automation (ICRA) 2018. (technical report)

2. M. Wei, V. Isler, A Log-Approximation for Coverage Path Planning with the Energy Constraint, International Conference on Automated Planning and Scheduling 2018. 

3. M. Wei, V. Isler, Air-To-Ground Collaboration for Energy-efficient Path Planning for Ground Robots, IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2019. (link)

4. M. Wei, V. Isler, Energy-efficient Path Planning for Ground Robots by Combining Air and Ground Measurements, Conference on Robot Learning 2019. (link)

5. M. Wei, V. Isler, Building Energy-Cost Map from Aerial Images and Ground Robot Measurements with Semi-supervised Deep Learning, IEEE Robotics and Automation Letters  (RAL), 2020. (link)

6. M. Wei, D. Lee, V. Isler, and Daniel. D. Lee, ‘Occupancy Map Inpainting for Online Robot Navigation’, IEEE International Conference on Robotics and Automation (ICRA), 2021 (link)

7. M. Wei, V. Isler, Predicting Energy Consumption of Ground Robots on General Terrains, IEEE Robotics and Automation Letters (RAL),  2021.

8. C. Peng, M. Wei, and V. Isler, ‘Stochastic Travelling Salesperson Problem with Neighborhoods for Object Detections’, IEEE International Conference on Robotics and Automation (ICRA), 2023.

Demos

navigateThroughDoor.MOV

Indoor  navigation

demopicking.mp4

Robot moving and picking

cornfiled0614Speed4x.mp4

     Cornfield with an autonomous robot

finalVideoSubmission.mp4

ICRA 2018 presentation video

pickingForceControl.mp4

Fruit picking with a force sensor to avoid squash

corl2019video_final_sub.mp4

CORL 2019 presentation video

Teaching

Florida Atlantic University

University of Minnesota