▪ Developed and curated a customized dataset of 500 images for training and validating an object detection model.
▪ Refined and translated a C++ pipeline to Python for robotic laser detection, leveraging OpenCV and for enhanced image processing capabilities.
▪Integrated production-level Object detection models like YOLOv8 to achieve precise center-finding, significantly enhancing the accuracy of robotic placements in subsequent operations.
▪Executed various vision tasks such as gauging, barcode reading, OCR, defect detection, part verification, pattern matching, sorting, Flexibowl picking, and 3D inspections.
▪Conducted feasibility studies and prototyped, commissioned, and evaluated machine vision systems for 8 companies.
▪ Created web interface using React framework integrated with ROS to monitor the robots.
▪ Developed computer vision pipeline for object detection and pose estimation of the robot.
▪ Research work focuses on an Ensemble Learning Methodology combining multiple existing robotics grasp synthesis algorithms to obtain comparatively better grasp quality.
▪ Train individual experts like Generative grasping, Residual-Net on Cornell dataset. Compare grasping performance of Ensembles architecture of these experts by evaluating diversity metrics.
▪ Conduct office hours and lab sessions. Grade assignments for 5 courses (Applied Statistics, Calculus IV, Geometric Concepts, Probability and Discrete Mathematics).
▪ Designed using Solidworks and manufactured an Automated Spool transfer machine using rotary encoder.
▪ Designed mechanical assembly setup for Spent Fuel Pool for temperature and level measurement with PLC system.
▪ Modelled rotary MR (Magneto-rheological) damper for Micro-actuator with multiple electromagnet coils.
▪ Conducted FE (Finite Element) based electromagnet analysis of the electromagnet on the proposed design.
▪ Performed CFD (Computational Fluid Dynamics) analysis on non-Newtonian fluid (Magneto-rheological fluid).
▪ Worked on “Study of Bumper defects and improvement of DOK (Daily efficiency)”. Around 4% DOK improvement was observed from the proposed method.
▪ Assisted with the setup for automation of Under-Body Coat Paint.
PyTorch | OpenCV | Numpy
▪ Developed unified encoder-decoder architecture to perform Depth and Surface Normalization with Semantic Seg-
mentation to boost performance on NYU2D dataset.
▪ Performed experiments using different weighting techniques (Grid search, Uncertainty weighting). Grid Search
offering good trade-off as MIOU increased from 0.751 to 0.806.
Open3d | YoloV4 | PyTorch | OpenCV | Numpy
▪ Calibrated camera using checkerboard pattern by mapping 3D world coordinates to image coordinates and then estimated parameters of camera through the processes of calculating intrinsic and extrinsic parameters.
▪ Implemented low-level (Point-Pixel) fusion and mid-level (Box to Box) fusion of LiDAR 3D points with 2D image detections using pretrained YOLOv4 network.
Tensorflow | Numpy
▪ Built a Time-series regression model with regularization techniques to predict crime rate for succeeding years.
▪ Implemented multiple classification (SVM, KNN, Decision Tree, SoftMax) and regression models for homicide
prediction, used Ensemble learning method with Max-Voting technique on classification methods and obtained an
accuracy of 88.92% .
Tensorflow | OpenCV | Numpy | Keras | Arduino UNO
▪ Created a Sequential model from dataset ”Plant Village”, which was trained on google colab. Resulted in 83%
accuracy and distinguished plants as healthy or diseased.
▪ Developed an IoT (Internet of Things) based smart agriculture system with moisture and LM35 temperature sensor
for monitoring soil environment.
Python | Pygame | Rviz | ROS | Gazebo
▪ Implemented Artificial Potential Field (APF) and Generalized Velocity Obstacle as local planner and RRT as global planner for dynamic obstacle avoidance on pygame and TurtleBot3 ROS Gazebo simulation.
Python
▪ Found feasible paths and compared distance metrics using discrete and sampling-based algorithms such as BFS, DFS, A*, Dijkstra’s, PRM, RRT and RRT*.
Solidworks | MATLAB | Simulink
▪ Design and fabricated 2 connected 1 Degree of Freedom robot using various spring mechanism.
▪ Optimized the design parameters for the system on MATLAB Simulink and provided the required torque using an
Admittance Sliding controller to the exo-skeleton system.
Solidworks | Arduino UNO
▪ Design using Solidworks and build a 3-DOF system with 2 cameras to track a tube inserted into a human body.
▪ Control the motors using Arduino to position the camera along with the movement of the tube.
Solidworks
▪ Worked in the chassis subsystem on Hybrid-electric and combustion prototype. Promoted to chassis module head
and led team of 5 people.
▪ Designed using Solidworks and fabricated fixture assembly for series-hybrid electric prototype.
▪ Developed a compact battery box design using Solidworks for stacking Lithium-ion batteries.
Robotics | Mechanical
Hey, I'm Chinmaya, a robotics master's student at the Worcester Polytechnic Institute. My focus area are at the intersection of Computer vision, Deep Learning and Machine Learning.
