Robotics Engineer with Vision Expertise
Uploaded by
gopivamsigiga123Robotics Engineer with Vision Expertise
Uploaded by
gopivamsigiga123Common questions
Powered by AIAt Calvary Robotics, Chinmaya Khamesra executed vision tasks such as barcode reading and Optical Character Recognition (OCR), among others, which are critical for verifying parts and ensuring accuracy in industrial operations. His integration of object detection models like YOLOv8 also improved precision in robotic placements, contributing to enhanced operational efficiency and accuracy .
At Denali Advance Integration, Chinmaya Khamesra developed a pipeline using YOLO and a RealSense camera to detect non-conveyable boxes in the Bulk-Sort flow, which likely improved the efficiency and accuracy of sorting processes in logistics settings. Additionally, he developed vision systems such as scan tunnels and sorting solutions for vendors including Amazon warehouses and FedEx, enhancing operational capabilities in logistics and warehousing .
Chinmaya Khamesra's research focused on an Ensemble Learning Methodology that combined multiple robotics grasp synthesis algorithms to enhance grasp quality. He trained individual experts like Generative grasping and Residual-Net using the Cornell dataset and evaluated the ensemble architecture's performance through diversity metrics, likely resulting in improved grasp synthesis outcomes compared to individual methods .
Chinmaya Khamesra's project on Multi-Task Self-Supervised Learning utilized a unified encoder-decoder architecture to perform Depth Estimation and Surface Normalization alongside Semantic Segmentation, achieving a performance boost on the NYU2D dataset. By applying different weighting techniques such as Grid search and Uncertainty weighting, the Mean Intersection over Union (MIOU) was increased from 0.751 to 0.806, resulting in a 7.3% improvement, demonstrating significant advancements in image segmentation tasks .
Chinmaya Khamesra enhanced robotics path planning on the WPI Campus Map by employing and comparing discrete and sampling-based algorithms such as BFS, DFS, A*, Dijkstra’s, PRM, RRT, and RRT*. This comprehensive comparison allowed him to evaluate the efficiency and practicality of different path planning techniques in finding feasible routes, improving path optimization on campus .
Chinmaya Khamesra integrated LiDAR and camera sensors by calibrating the camera using a checkerboard pattern to map 3D world coordinates to image coordinates. He then implemented low-level (point-pixel) and mid-level (box to box) sensor fusion using a pretrained YOLOv4 network. This integration allowed for more robust 3D point and 2D image detection, enhancing machine vision insights for complex spatial analyses .
Chinmaya Khamesra's development of a web interface using the React framework integrated with ROS significantly impacted robotic operations by facilitating real-time monitoring and control. This interface likely improved user interaction and operational efficiency by enabling seamless human-machine communication, vital for responsive and adaptable robotic systems .
Chinmaya Khamesra developed synthetic good images using DCGANs and evaluated EfficientAD models for anomaly detection, comparing results on original and synthetic datasets. He also led the setup and integration of vision systems for battery module production lines and created a web interface for real-time camera-HMI system communication, advancing automation and efficiency in production processes .
Chinmaya Khamesra used a CNN-based Sequential model and IoT technology to classify leaves as healthy or diseased, achieving an 83% accuracy rate. This approach was effective in providing a reliable automated tool for monitoring plant health with applications in smart agriculture systems .
Chinmaya Khamesra faced challenges in accurately predicting crime rates for future years through Time-series regression and regularization techniques while maintaining sufficient model accuracy. His achievements included implementing multiple classification models such as SVM, KNN, Decision Tree, and SoftMax for homicide prediction, ultimately achieving an 88.92% accuracy using an Ensemble learning method, which indicates high predictive performance amidst the complexities of crime data .