Research

In general, my research interests are robot motion planning, mobile robotics, software validation and verification, and computer science education. My research contributions are described below.

Jump to Publications

Real-time Adaptive Motion Planning (RAMP)

Robot motion in practical environments is still a challenge despite decades of research. My research on this topic focuses around utilizing the RAMP approach to navigate a non-holonomic robot at real-time to reach a goal in a dynamic environment containing unknown obstacles that move in unforeseen and arbitrary ways.

RAMP for Non-holonomic Motion

The vast majority of mobile robots are subject to non-holonomic motion. Therefore, it is imperative that motion planning frameworks are capable of producing motion under these constraints. However, these constraints are not trivial to plan around.

My research was focused on incorporating non-holonomic constraints into the RAMP framework. After achieving this, a real-time perception algorithm was developed to allow RAMP to work in the presence of unknown obstacles.

Specifically, my contributions to the approach are:

  • A hybrid trajectory representation that contains both non-holonomic and holonomic segments
  • A method for real-time trajectory switching using non-holonomic trajectories
  • Adaptive control cycles based on a robot’s dynamics
  • A straight-forward method for uncertainty reduction based on cycle interactions in the RAMP framework
  • A method to quickly approximate obstacle shapes to detect unknown obstacles and predict unforeseen obstacle motion.

-

  • Sterling McLeod (2019). Robust and reliable real-time adaptive motion planning. ProQuest LLC. Link bibtex

  • Sterling McLeod and Jing Xiao. Real-time adaptive non-holonomic motion planning in unforeseen dynamic environments. In IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2016. DOI bibtex

Learning with RAMP

Two approaches to incorporate learning into the RAMP framework have been studied.

Hilbert maps

Nearly all robot planning frameworks guide a robot through an environment as if it never navigated the environment before; no past experience is incorporated. However, a person navigates an environment utilizing past experience even though the environment has new dynamic obstacles with unknown motion each time the person visits.

For example, inside a food court, a shopper navigates among many unknown people with unknown motion (i.e., without knowing where people are going) but can still utilize past experience (for instance, chairs at a table are likely to be pulled away from the table when people sit down) to navigate more efficiently without collision in such an environment.

The work presented in this paper is inspired by the above observation and combines learning from past experience and real-time motion planning.

-

  • Sterling McLeod and Jing Xiao. Navigating Dynamically Unknown Environments Leveraging Past Experience. In IEEE International Conference on Robotics and Automation (ICRA), 2019. DOI bibtex

Reinforcement learning

Real-time motion planning of robots often needs to consider multiple and sometimes conflicting optimization criteria, such as time efficiency (in terms of the shortest distance or time), safety (in terms of the clearance to obstacles), and energy efficiency. RAMP combines these criteria in a cost function as a weighted sum. Determining proper values for the coefficients in the cost function is not a trivial issue, but it is often done manually in an ad hoc manner. Moreover, when the task environment changes, the previously set coefficient values may not be suitable anymore.

The goal of our work in this area was to develop a learning framework that can change the cost function coefficients on the fly based on the current environment.

  • Kai Zhang, Sterling McLeod, Minwoo Lee, Jing Xiao. Continuous reinforcement learning to adapt multi-objective optimization online for robot motion. International Journal of Advanced Robotic Systems. March 2020. DOI bibtex

Model-based Test Generation for RAMP

Real-world environments are dynamic, unpredictable, and large. It is impossible to generate an exhaustive list of test scenarios for a motion planning algorithm. However, we still desire a way to validate a software system implementing a motion planning algorithm for dynamic environments with obstacles that move in unforeseen arbitrary ways.

My work in this area has focused on System-level testing of a RAMP system. The papers linked [4,5] below shows a use-case of applying a model-based approach to RAMP for Component-Integration testing, and a novel method to generate System-level tests for RAMP.

Alt

  • Mahmoud Abdelgawad, Sterling McLeod, Anneliese Andrews, and Jing Xiao. Model-based testing of a real-time adaptive motion planning system. Advanced Robotics, 31(22):1159–1176, 2017. DOI bibtex

  • Mahmoud Abdelgawad, Sterling McLeod, Anneliese Andrews, Jing Xiao. Model-Based Testing of Real-Time Adaptive Motion Planning (RAMP). In IEEE International Conference on Simulation, Modeling, and Programming for Autonomous Robots (SIMPAR), 2016. DOI bibtex

Publications

  1. Kai Zhang, Sterling McLeod, Minwoo Lee, Jing Xiao. Continuous reinforcement learning to adapt multi-objective optimization online for robot motion. International Journal of Advanced Robotic Systems. March 2020. DOI bibtex

  2. Sterling McLeod (2019). Robust and reliable real-time adaptive motion planning. ProQuest LLC. Link bibtex

  3. Sterling McLeod and Jing Xiao. Navigating Dynamically Unknown Environments Leveraging Past Experience. In IEEE International Conference on Robotics and Automation (ICRA), 2019. DOI bibtex

  4. Mahmoud Abdelgawad, Sterling McLeod, Anneliese Andrews, and Jing Xiao. Model-based testing of a real-time adaptive motion planning system. Advanced Robotics, 31(22):1159–1176, 2017. DOI bibtex

  5. Mahmoud Abdelgawad, Sterling McLeod, Anneliese Andrews, Jing Xiao. Model-Based Testing of Real-Time Adaptive Motion Planning (RAMP). In IEEE International Conference on Simulation, Modeling, and Programming for Autonomous Robots (SIMPAR), 2016. DOI bibtex

  6. Sterling McLeod and Jing Xiao. Real-time adaptive non-holonomic motion planning in unforeseen dynamic environments. In IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2016. DOI bibtex