Roshan Dhakal

I'm a Ph.D. candidate researching at the intersection of Robotics and Machine Learning in Robotic Anticipatory Intelligence & Learning (RAIL) Lab advised by Dr. Greg Stein at George Mason University (GMU), Fairfax, VA. I received my Masters degree in Computer Science from GMU in 2021. I completed Bachelors degree in Computer Engineering from Tribhuvan University, Nepal.

Feel free to contact me at rdhakal2@gmu.edu or dhakalrosan@gmail.com

Email  /  Resume  /  Research Statement  /  Scholar  /  Github

profile photo

News

  • (November 2025) I was awarded with Dissertation Completion Grant from Provost Office, George Mason University.
  • (October 2025) Our paper on Anticipatory Task and Motion Planning got accepted at IEEE RA-L 2025.
  • (May 2025) I was awarded with Provost Scholarship at George Mason University.
  • (November 2024) I was recognized as outstanding reviewer at CoRL LEAP 2024..
  • (June 2024) I successfully defended my Ph.D. proposal and advanced to candidacy.
  • (May 2023) I presented Anticipatory Planning paper in ICRA 2023.
  • (January 2023) Our paper on Anticipatory Planning got accepted at ICRA 2023.
  • (May 2021) Successfully defended my Ph.D. comprehensive exam.

Research

I envision that household and service robots will live with humans in near future. My research aims to understand not only how a robot accomplish a complex task but also on how its actions can impact the future despite missing knowledge of the environment it lives in. I am motivated by the challenge of improving a robot's performance in long-lived task and motion planning settings by enabling robots to anticipate the impact of their actions on the environment and act intelligently.

My research focuses on task and motion planning, learning augmented planning and lifelong robot planning. Below are some of my works.

anttamp Anticipatory Task and Motion Planning: Improved Rearrangement in Persistent Continuous-Space Environments.
Roshan Dhakal, Duc M. Nguyen, Tom Silver, Xuesu Xiao, Gregory J. Stein
IEEE Robotics and Automation Letter (RA-L), 2025
bibtex

We present a learning-guided task and motion planning that enables a long-lived robot with forward thinking behaviors while difficult to achieve otherwise—organization, tidiness and preparation.

ant-task-plan Anticipatory Planning: Improving Long-Lived Planning by Estimating Expected Cost of Future Tasks.
Roshan Dhakal, Md Ridwan Hossain Talukder, Gregory J. Stein
IEEE ICRA, 2023
video / bibtex

We propose anticipatory task planning to enable a robot consider unseen future tasks while solving a given task.

gnn-room-classif Room Classification on Floor Plan Graphs using Graph Neural Networks.
Abhishek Paudel, Roshan Dhakal, Sakshat Bhattari
arxiv, 2021
bibtex / code

We present our approach to improve room classification task on floor plan maps of buildings by representing floor plans as undirected graphs and leveraging graph neural networks to predict the room categories.

Projects

Below are some of the side projects I have worked on.
NAMO Example Integrating learned cost estimator as heuristic in Fast Downward Planning System using Pybind and C++

We augment the planner’s search with a pytorch trained learned heuristic that composes with standard ones (e.g., FF, max, goal-count).

NAMO Example Navigation among movable obstacles using PDDLStream

In a simple 2D setup, we encode the domain in PDDL and solve it with PDDLStream to relocate obstacles and open a path to the goal.

Lunar Lander Comparison between reinforcement learning algorithms on OpenAI gym environment.

I compare approaches on solving the lunar lander problem in a open AI gym using two reinforcement learning methods; Proximal Policy Optimization (PPO) and Deep Q Network (DQN).


Stole website template from Jon Barron source code.