Intelligent Automation for Critical Real-World Systems
We investigate theoretical foundations and deploy physical robotic systems across commercial greenhouses, advanced manufacturing lines, and clinical rehabilitation centres.
GIGAS: Intelligent High-Wire Greenhouse Harvesting Robot
The Guelph Intelligent Greenhouse Automation System (GIGAS) is an intelligent robotic platform engineered for high-wire greenhouse tomato harvesting. Evaluates pedicel-targeting hold-and-cut mechanisms (P1, P2, S1, S2) alongside multi-modal tomato-targeting soft grippers (SG1, SG2) to overcome cluster occlusions and delicate fruit handling. The flagship SG2 end effector combines pneumatic four-bar linkage fingers (38.8 mm to 104 mm span), a double-bevel vacuum suction cup (7.8 N/cm² suction), and closed-loop force-torque sensing to locate and detach mature beefsteak tomatoes at the natural pedicel abscission zone. Demonstrated 100% positioning success and an 81% overall harvesting success rate across commercial greenhouse trials at AMCO Farms (Leamington, ON) and Singh Greenhouses (Millgrove, ON) without produce bruising or calyx damage.
Automated In-Line Inspection Station for Automotive Gears
An industrial automated vision inspection station designed for 100% inline quality control of automotive pinions and gears. Solves high-throughput manufacturing cycle time constraints by optimizing stepper actuator kinematics, high-speed camera exposure triggers, and a multi-threaded asynchronous control server architecture. Reduced scan and inference cycle time from 90 seconds to 7.5 seconds per pinion while validating defect detection on 537 scanned gears in production environments.
Head-Controlled Interface for Human Assistive Robotic Devices
A hands-free, vision-guided assistive feeding platform designed for individuals with severe upper-limb motor impairments. Integrates a YOLOv9-Seg network for food item and modular tray instance segmentation, MediaPipe Face Mesh tracking for intuitive head-gesture navigation, and a mouth-open dwell-free confirmation trigger. Evaluated in user trials with 22 participants, demonstrating 84.7% head-tracking accuracy, 92.6% graph traversability, and a 9.69/10 safety rating on commodity CPU hardware.
Multi-Stage Deep Learning Pipeline for Automated Weed Scouting in Lima Beans
An automated field-scouting computer vision system for whole-field weed density estimation in commercial lima bean crops, targeting problematic species such as Nightshade and Pigweed. Deploys a three-stage deep learning pipeline combining broad-scale crop-versus-weed object detection with fine-grained species classification and GIS density mapping. Evaluated across three consecutive field seasons (2023–2025) in Southwestern Ontario in partnership with Haggerty Agrobotics and Nortera Foods, achieving 76.35% detection precision.
Virtual Reality Mirror Therapy System for Stroke & ABI Rehabilitation
An immersive Virtual Mirror Therapy (VMT) neurorehabilitation system designed to treat upper-extremity hemiplegia following stroke and acquired brain injury (ABI). Replaces conventional physical mirrors with an Oculus Rift CV1 VR headset, Leap Motion optical hand/finger tracking, and Microsoft Kinect v2 skeletal tracking to create an interactive virtual environment that projects real-time mirrored bilateral movements of the unimpaired limb onto the hemiplegic limb avatar. Clinically evaluated in a randomized controlled trial (RCT) with 7 inpatients at the Hamilton Health Sciences Regional Rehabilitation Centre across five functional rehabilitation activities, demonstrating significantly higher patient engagement (16.7 vs 6.0 on RTES) and increased therapy repetition rates compared to traditional mirror therapy.
Robotic Bin-Picking of Large Deformable Automotive Components
An intelligent robotic bin-picking system designed to unpack and sort large, semi-rigid Expanded Polypropylene (EPP) automotive components from highly cluttered and occluded conveyor bins. Features a novel compliant dynamic vacuum end-effector that accommodates sensor noise and part deformation, combined with 3D point cloud surface segmentation algorithms to reliably detect graspable regions across 8 distinct industrial part geometries.
GRETA: Guelph Robot for Everyday Task Assistance
A personal care assistive robotic system designed to restore independence in hygiene Activities of Daily Living (ADLs) for individuals with upper-limb motor impairments. Features quick-swap instrumented end effectors coupled with a Robotiq FT 150 6-DoF force/torque sensor and Microsoft Kinect vision tracking for personalized 3D facial tool-path planning. Evaluates static and dynamic contact force profiles across facial, cranial, and neck regions during hair brushing, face wiping, and shaving tasks to establish safety limits and adaptive force-guided compliance algorithms.