Abdullah Khaled

BU RISE: Benchmarking Reinforcement Learning Algorithms with Search and Rescue Tasks

As part of the Boston University Research in Science and Engineering program, I was mentored by Dr. Wenchao Li and Zijian Guo. Specification-guided (spec-guided) reinforcement learning (RL) attempts to improve the performance of traditional RL algorithms in long-horizon tasks, or tasks that require goals to be met in a certain order. Search and rescue (SAR) is one such task, as agents often must complete rescues in a specified order (e.g. based on risk level). In this work, we propose an extension to existing spec-guided RL benchmarks that includes a series of long-horizon SAR tasks compatible with spec-guided and traditional RL methods. We trained unconstrained, constrained, and spec-guided RL models in our multi-agent environments. Empirical results find that spec-guided RL model outperforms the unconstrained and constrained traditional models in the SAR environments!
Read More

FIRST Robotics Competition: 2026 Season

The 2026 Season of FRC, REBUILT, consisted of three robots on one alliance scoring FUEL into a HUB during their respective shifts. During this season, my team had (in my opinion) the most impressive year in our history, ending as the Winner of both district events we competed in, Impact Award Winner of our second district event, Winner of the Texas State Championship, Semi-Finalist of World Championship Newton Division playoffs, and Division Engineering Inspiration Award winner! As Technical Advisor, I oversaw our software development on the robot. I guided other software members through creating and refining our subsystems while developing more complex functions including pose estimation, auto-align, and autonomous routines. By the end of the season, we (statistically) reached 19th in the world rankings and 4th most improved team since last year!
Read More

OralVision: Federated Machine Learning Intraoral Screening System

Oral cavity cancers are highly treatable when caught early, but almost 70% of cases are diagnosed too late due to clinical uncertainty and a lack of accessible screening tools, leading to over 170,000 deaths annually. OralVision is an intraoral, deep-learning-based diagnostic device designed to change this unfortunate reality. By integrating noninvasive imaging with advanced artificial intelligence, OralVision could provide physicians with highly accurate diagnostic insights, allowing for earlier clinical referrals and therapeutic intervention. Moreover, with its use of inexpensive yet powerful components, we can mass produce and distribute OralVision at a much lower marginal cost than existing solutions, contributing to improved 5-year survival rates, especially in marginalized low-income communities.
Read More

Mantik: A FIRST Robotics-Centric Java Programming Curriculum

After completing the 2025 REEFSCAPE season as the only programmer on my team, I wanted to make a resource to institutionalize the knowledge I gained during my time in FRC and help build a deeper bench of programmers to prevent burnout in future seasons. Therefore, I created Mantik, a website that trains students in FIRST® Robotics programming concepts ranging from basic motor to advanced command-based libraries and control theory. By combining the documentation of a variety of sources, from the official WPILib documentation to each vendor's specific APIs, I was able to create accurate, concise tutorials for these complex concepts, which I published to my team as well as others. In addition, I created a series of video tutorials aimed at guiding students through these difficult concepts.
Read More