Matthew Lisondra

I am currently a Researcher of Robotics at the University of Toronto Robotics Institute.

My research focuses on Robot Perception, Robot Learning, Computer Vision, Simultaneous Localization and Mapping (SLAM), Autonomous/Intelligent Systems Algorithms, High Framerate Processing Low-Power Unconventional Sensing, 3D Scene Representations and Embodied AI.

I am affiliated with the following research labs:

• Autonomous Systems and Biomechatronics Lab (ASB) directed by Dr. G. Nejat
• Robotics and Computer Vision Laboratory (RCVL) directed by Dr. S. Saeedi
• Haptics and Telerobotics Laboratory (HapTel) directed by Dr. K. Zareinia
• Ingenuity Labs Research Institute (with Dr. A. Wu)

Email  /  Curriculum Vitae (CV)  /  LinkedIn  /  Google Scholar

UofT Robotics Institute, University of Toronto PhD, Doctor of Philosophy (Mechanical Engineering)

• Focus: Robot Perception, Robot Learning, Computer Vision
• Leveraging LLM and VLM agents w/ Computer Vision for Robot Learning
• Research at the University of Toronto Robotics Institute
• Supervised by: Dr. G. Nejat of the ASB Lab

Robotics and Computer Vision Laboratory (RCVL) MASc, Master’s of Applied Science (Mechanical Engineering)

• Focus: Computer Vision and Visual-Inertial SLAM
• Designed the first 6-DOF Visual Inertial Odometry on FPSPs (BIT-VIO)
• Thesis Published, Presentation in Yokohama, Japan for IEEE 2024 ICRA
• Collaboration with Imperial College London and University of Manchester
• Supervised by: Dr. S. Saeedi of RCVL, Dr. K. Zareinia of HapTel Lab

Physics, University of Toronto HBSc, Honours Bachelor of Science (Physics and Computer Science)

• Focus: Robotic Mechanics, Probability, TS-Analysis, Computational Physics
• Research: Time Series Analysis on Global Temperature, Sea Level Pressure
• Research: Helium-Neon Laser Analysis (Reviewed by Dr. A. Vutha)
• Research: Percolation via Random Processes Monte Carlo, Porous Rock
• Collaborated with: Dr. D. Jones of APCM Group

Presents the first systematic review of foundation models integrated into mobile service robotics
Identifies and categorizes the core challenges in embodied AI: multimodal sensor fusion, real-time decision-making, task generalization, and effective human-robot interaction
Analyzes how foundation models (LLMs, VLMs, MLLMs, VLAs) enable real-time sensor fusion, language-conditioned control, and adaptive task execution
Examines real-world deployments across domestic assistance, healthcare, and service automation domains
Demonstrates the transformative role of foundation models in enabling scalable, adaptable, and semantically-grounded robot behavior
Proposes future research directions emphasizing predictive scaling laws, autonomous long-term adaptation, and cross-embodiment generalization
Highlights the need for robust and efficient deployment of foundation models in human-centric robotic systems

Designed the first tightly-coupled 6-DoF Visual-Inertial Odometry for FPSPs (TCB-VIO)
Operates at 250 FPS with on-sensor binary edge input and 400 Hz IMU integration via MSCKF
Introduced binary-enhanced KLT tracker for robust and smooth feature tracking on SCAMP-5
Reduces latency and power via direct on-sensor SIMD processing with limited pixel memory
Outperforms VINS-Mono, ORB-SLAM3, and ROVIO under aggressive motion and fast dynamics
Achieves long, stable feature tracks in real-world tests with motion-capture ground truth
Extends OpenVINS tightly-coupled architecture for binary descriptors and high-speed tracking

Developed a high-frequency ultrasound-based method for microplastic detection and sizing
Designed a peak extraction technique to isolate strong backscatter signals for analysis
Achieved 99.63% material classification accuracy using a custom 1D-CNN (PE, PMMA, steel, glass)
Performed size estimation with a material-specific MLP achieving 99.97% accuracy
Combined classification and sizing into an integrated machine learning pipeline
Provides a fast, low-cost, and automated alternative to traditional spectroscopic methods
Validated on a custom acoustic dataset, enabling potential for real-time in-situ monitoring

Developed the first multimodal deep learning system for multi-activity ADL recognition in SARs
Introduced ADL and motion embedding spaces to detect known, unseen, and atypical ADLs
Proposed a user state estimation method using similarity functions for real-time ADL inference
Enabled proactive HRI by allowing SARs to autonomously initiate context-aware assistive behaviors
Overcame false positives and limited generalization in prior ADL recognition systems
Achieved higher classification accuracy than state-of-the-art human activity recognition models
Validated in cluttered home-like environments using the socially assistive robot Leia
Adapted to diverse users by recognizing variable and non-standard ADL execution styles

Presents a novel technique capable of generating geometric maps from WiFi signals
A novel algorithm that is capable of generating geometric maps using WiFi signals received from multiple routers
Benchmarking the WiFi-generated maps with Lidar-generated maps by comparing the area,
number of data points, RSSI prediction True/False setting, RSSI accuracy percentage, IOU and MSE scores
Evaluation on real-world collected from indoor spaces

Designed the first 6-DOF Visual Inertial Odometry on FPSPs (BIT-VIO)
Efficient VIO operating and correcting by loosely-coupled sensor-fusion iEKF at 300 FPS
using predictions from IMU measurements obtained at 400 Hz
Uncertainty propagation for BIT-VO's pose as it is based on binary-edge-based descriptor extraction
Extensive real-world comparison against BIT-VO, with ground-truth obtained using a motion capture system

Studied the usability and advantages of FPSPs to leverage a more accurate state estimation framework
Designed an algorithm for VIO using Focal-Plane Binary Features
Implemented the FPSP vision- IMU-fused estimation algorithm on a mobile device for offline and online real-world testing
Evaluate the performance, benchmarking against FPSP vision-alone and ground-truth data
Extensive study on the algorithmic execution timing/frame, accuracy, memory usage
and power consumption of the visual front-end processing performance on the FPSP

• Reviewer (Conference) for IROS 2025, IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)
• Reviewer (Journal) for ISJ 2025, IEEE Sensors Journal (ISJ) 2025
• Reviewer (Journal) for RA-L 2024, IEEE Robotics and Automation Letters (RA-L) 2024
• Reviewer (Conference) for IROS 2024, IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)
• Reviewer (Conference) for ICRA 2024, International Conference on Robotics and Automation (ICRA) 2024
• Reviewer (Conference) for IEEE CCECE 2023, 2023 Canadian Conference On Electrical and Computer Engineering
• Reviewer (Conference) for IROS 2023, IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

I am affiliated with several teaching institutions, teaching Physics (Advanced) and Computer Science. Select students I have taught:

• H. Li
• L. Liang
• N. Feng
• R. Sun* (Multiple Scholarship Offers in Australia, England)
• W. Zheng
• A. Zhang
• A Huang* (Now at Berkeley Music School)
• B. Zheng
• B. Luo* (Full Scholarship Physics, University College London)
• B. Yang* (Now at King's College London)
• C. Yang* (Now at Durham University)
• F. Chen
• G. Ye* (Now at University of Hong Kong)
• J. Wu
• K. Sheng, Chuwen (Multiple University Offers)
• L. Lin
• N. Zhao* (Multiple University Offers in Business)
• N. Tian
• A. Pu
• D. Qiu
• M. Wang
• R. Zhang
• C. Zou
• Y. Caelon
• S. Yumo
• A. Ma
• A. Xu
• K. Chen
• F. Xin
• A. Yi Lan
• C. Yi
• F. Chen
• L. Li
• M. Di
• R. Zhang
• S. Fan

Thesis Mentoring/Guidance: Junseo Kim (Now at TUDelft Robotics)

Autonomous Truck Navigation with Trailer Integration via Natural Language Processing (NLP)

More information as well as students taught can be found on my Curriculum Vitae (CV).

Rosor Exploration

• Researcher - Robotics, Geoscientific UAVs and Drones
• Working on Development of Rosor's Active Terrain Following (ATR) System
• Currently: Now Led by H. A. Jafaar

Outside of Research, I do Filmmaking with Film Director Dinitha Vithanage and Film Editor Brandon Worden. We just recently presented our Feature-Film.