Machine Learning Engineer (Perception, Localization, and Mapping) @ Dematic, Kion Mobile Automation

• Developing intelligent, automated solutions for manufacturing, warehouse, and distribution environments, sold by KION Group and its operating units Dematic, STILL and Linde.
• Directly involved with developing on-vehicle Simultaneous Localization and Mapping (SLAM) software for AGVs and AMRs. Also contributing to automated driving algorithms for collaborative AGVs performing order picking.
• Improved accuracy and robustness of the reflector extraction algorithm leading to sub-centimeter accuracy.
• Developed a Gazebo-based virtual testing pipeline to improve line extraction algorithm using hyperparameter tuning.
• Improved speed of feature-based association algorithm by 50%, leading to a 30% faster localization pipeline. Analyzed real-time CPU utilization of SLAM processes and threads using Valgrind, LTTng, and perf.
• Standardized the pallet pick/drop testing at physical warehouses. The accuracy-repeatability analysis was done using an external laser tracking system by Faro. Also, a procedure to calibrate lidar, steering encoder, and traction encoder was created.
• Developed a python-based tool to visualize recorded SLAM logs, and automated their offline performance analysis.
• Evaluated the use of Visual Inertial Odometry and GraphSLAM to improve AGV navigation in warehouse environments.
• Certified SAFe (Scaled Agile Framework) Practitioner and trained to use Scrum, Kanban, and XP in a SAFe environment.
• To see the AGVs in action at customer sites, check out the videos here, here, here and here.

Robotics Software Engineer Nanodegree Program @ Udacity

• Localization, Mapping, and Path Planning for Autonomous Vehicles in Factories (repo)

  • Simulated Automated Guided vehicles (AGVs) and Autonomous Mobile Robots (AMRs) in a warehouse environment.
  • Mapped a virtual environment, by manually moving an AGV and creating an occupancy grid map. The AGV only had a 2D lidar sensor on it. Localized an AMR using the same map while controlling it manually or autonomously using path planning. Simulated a complete pick and drop operation. ROS packages like gmapping, AMCL, and move_base were used.
  • Used SLAM to manually move an AMR through a virtual environment and create a loop-closed 3D graph map using RTAB-Map. The AMR used a 3D camera and a 2D lidar sensor to create the map for localization.
  • Programmed an AMR to autonomously follow a moving reflective ball target using a 2D camera.

• Path Planning for KUKA robotic arm (repo)

  • Used A-star path planning to move a payload through a 2D maze using a KUKA industrial robotic arm as part of KUKA Robotics Challenge.

Self-Driving Car Engineer Nanodegree Program @ Udacity

• Lane Detection (repo1), (repo2)

  • Created a robust image processing pipeline to detect a highway lane in an image, pre-recorded video, or live feed from a dashcam.
  • Calculated the car’s position within lane and lane curvature using perspective transform and polynomial fitting.

• Traffic Sign Classification (repo)

  • Developed a LeNet inspired convolution neural network using TensorFlow to classify the GTSRB traffic sign dataset.
  • Achieved 94.8% accuracy on test dataset by data augmentation and image enhancement using OpenCV.

• Traffic Light Awareness

  • Used SSD Mobilenet for real-time traffic light detection and classification.

• Behavioral Cloning (repo)

  • Designed an end-to-end convolution neural network using Keras that predicts steering angles from dash-cam images.
  • Successfully cloned human driving behavior to autonomously steer a car around a virtual test track after neural network tuning and data augmentation.

• Sensor Fusion (repo)

  • Implemented car location estimation algorithm using extended Kalman Filter based on LIDAR and RADAR sensor data.

• Localization (repo)

  • A 2D particle filter for sparse localization is designed and uses GPS and sensor data with a landmark map.

• Trajectory Planning (repo)

  • A Finite State Machine based planner is created to achieve autonomous highway driving with other cars.
  • Jerk minimized trajectories are considered to enhance occupant safety.

• Control (repo)

  • A PID controller is implemented to maneuver a vehicle around a virtual track using steering, throttle, and brake.

• System Integration

  • Robot Operation System (ROS) is used to robustly combine Perception, Planning, and Control.

Research Assistant @ CAD Innovation and Engineering Lab

• Robotics

  • Developed SnappyXO, a laser-cut design-driven robotics platform that enables designing mechanisms, structures, and robots. It has successfully raised $16K+ on Indiegogo for a crowdfunding campaign.
  • Organized workshop “Designing, Prototyping and Programming Robot Motions using MotionGen and SnappyXO” at IDETC 2018 and “Kinematic Summer School” at IDETC 2019.
  • Mentored multiple teams, as part of the Vertically Integrated Projects (VIP) program, for Robotics-driven Engineering Design Exploration. Some of the projects were Walking Robots, Self-driving Car, Sumo/Battle Bot, Biomimetic Robots, Drones, and Bots with Arduino alternative microcontrollers.
  • Assisted in developing and improving Rehabilitation Robots for sit-to-stand assist and gait correction.
  • Collaborated with ROAR lab at Columbia University on the design of spatial robots for neck rehabilitation.
  

  
  

• Mechanism Design Software (repo)

  • Developing a Computational Framework for Data-Driven Mechanism Design Innovation supported by $450K NSF grant.
  • Created MotionGen, a web-based mechanism design framework. Try it out here.
  • It uses MEAN (MongoDB, Express.js, Angular.js, Node.js) stack to create a RESTful web service based on MVC architecture.
  • Both iOS and Android apps have been created using Apache Cordova framework.
  

  
  

• Research and Algorithms

  • Path synthesis of mechanisms based on Fourier descriptor fitting using Nelder-Mead and Simulated Annealing optimization.
  • Mixed motion and path mechanism synthesis using optimal non-uniform DFT and Singular Value Decomposition.
  • Real-time simulation of planar, spherical, and spatial mechanisms with prismatic and revolute joints using Newton-Raphson optimization.
  • Developed unified planar, spherical, and spatial mechanisms synthesis techniques using Homotopy continuation method for simultaneous type and dimensional synthesis.
  • Developing machine learning based techniques for path synthesis using spatial platform linkages.
  
  
  

  
  

Product Design @ Vivonics, Inc.

• Coordinated with the design team on developing PMT Monitor, a portable medical headset that detects head trauma.
• Generated concepts for a mechanism that adjusts the interpupillary distance between the lenses focusing on manufacturability and robustness.

Design and Manufacturing @ Leviathan Energy

• Designed and manufactured a Hydro-kinetic turbine with an improved airfoil design that produces 50% more power in collaboration with Leviathan Energy.
• Created engineering models using Solidworks and Autodesk Inventor and fabricated parts by FDM based 3D printing.

Research Fellow @ IIIT Jabalpur

• Led a $70k+ research project funded by the Science and Engineering Research Board titled “Development of Additive-Subtractive Integrated Rapid Prototyping System for Improved Part Quality”.
• Spearheaded design and manufacturing teams to create a new hybrid 3D printing process using Pellet based Screw Extruder with CNC machines. Created Toolpath Planning strategies to manufacture CAD models using Hybrid Manufacturing techniques.

Design @ Sara Sae

• Involved with product development of Hydraulic Tongs used in the oil and gas industry.
• Carried out Kinematic and Dynamic analysis to find the most probable failure regions.