Parth Potdar Email & Phone Number

Oxford, Gb

London, Gb

City Of Bristol, England, United Kingdom

Research in bio-inspired aerial robotics in a small team, leading the design and development of anautonomous drone with a novel, tendon-actuated, passive perching mechanism for energy-efficientforest monitoring. An academic paper is in progress.Designed a compliant perching mechanism and contributed to computer vision, Visual-InertialSLAM, and drone electronics; trained a YOLOv8 model on custom images for branch detection andfurther image analysis for branch… Show more Research in bio-inspired aerial robotics in a small team, leading the design and development of anautonomous drone with a novel, tendon-actuated, passive perching mechanism for energy-efficientforest monitoring. An academic paper is in progress.Designed a compliant perching mechanism and contributed to computer vision, Visual-InertialSLAM, and drone electronics; trained a YOLOv8 model on custom images for branch detection andfurther image analysis for branch selection.Implemented ROS/ROS2 nodes and packages in Python for system control; utilized depth cam-eras (Intel RealSense D430i, D455) and VINS-MONO SLAM for drone localization and branchmeasurements.Integrated NVIDIA Jetson Orin Nano as an onboard computer, and worked with flight controllers(Pixhawk 6C, PX4 software) and drone electronics, including wiring, servo control, and housingdesign.Conducted successful perching tests on real tree branches; produced weekly progress presentationsand currently contributing to an academic paper. Show less

Cambridge, England, United Kingdom

Independent research on vision-based tactile perception for braille reading. Demonstrated a state-of-the-art speed of 315 words per minute at 87.5% accuracy, more than twice the speed of humanbraille reading. An academic research paper accepted for publishing in the IEEE RA-L journal 2024and cited multiple times. DOI: 10.1109/LRA.2024.3356978Designed a Braille-reading robot using a UR3 robotic arm with a DIGIT tactile sensor attachment,a SolidWorks-designed 3D-printed… Show more Independent research on vision-based tactile perception for braille reading. Demonstrated a state-of-the-art speed of 315 words per minute at 87.5% accuracy, more than twice the speed of humanbraille reading. An academic research paper accepted for publishing in the IEEE RA-L journal 2024and cited multiple times. DOI: 10.1109/LRA.2024.3356978Designed a Braille-reading robot using a UR3 robotic arm with a DIGIT tactile sensor attachment,a SolidWorks-designed 3D-printed adapter, and force measurements from a calibrated strain gaugevia Arduino.Developed a full Braille reading pipeline: frames captured with a tactile sensor, deblurred witha trained autoencoder, classified by a YOLOv8 model, and error-corrected using a data-drivenconsolidation process.Trained the autoencoder on synthetically blurred images and manually labelled detection data inPyTorch, due to lack of existing datasets; implemented confidence-based error correction to improveaccuracy.Initial research explored traditional CV algorithms (SIFT and Hough Transform), soft materialmodelling using MATLAB Simulink and Recurrent Neural Networks. Show less

Nottingham, England, United Kingdom

Completed 4 independent case studies with the goal of evaluating Reduced Order Modelling andMachine Learning (ML), and then creating a working prototype software to automate the process.The results of these studies were presented to a team of Project Managers at the end of myinternship.Gathered data and created ML models to predict the safety factors and damages of gears andbearings in a Romax gearbox model, with less than 1% error. Compared the proprietary ML… Show more Completed 4 independent case studies with the goal of evaluating Reduced Order Modelling andMachine Learning (ML), and then creating a working prototype software to automate the process.The results of these studies were presented to a team of Project Managers at the end of myinternship.Gathered data and created ML models to predict the safety factors and damages of gears andbearings in a Romax gearbox model, with less than 1% error. Compared the proprietary ML package,ODYSSEE, with Python’s Sci-kit Learn library, assessing the accuracy and run-time of both.Created Python and C++ scripts to communicate between Romax and ODYSSEE using a commandline interface, to automate data gathering and ML model creation. Trained fully connected NeuralNetworks (NN) with PyTorch.Improved upon the initial C++ applications by validating its performance and usability in thecreation of a durability model, which predicted component damages for any load case, for a gearboxmodel. Initial data analysis motivated the creation of a custom clustering algorithm. Show less

Master Of Engineering - Meng, Information And Computer Engineering, Predicted 1St

Project: Spatial uncertainty in navigation in the brain: simulating a Bayesian image-computable observer agent to infer a probability.

Bachelor'S Degree, Aerospace And Aerothermal Engineering, Mechanical Engineering, 2.1

Modules: State-space control, Digital signals and systems, Fluid mechanics, Dynamics, Vibrations, PDEs and variational methods, Business.

A Level

Subjects: Maths, Further Maths, Physics, Chemistry, and EPQ Grades: 5 A*s