Sanket Salunkhe

As a Robotics Research Scientist at the NYU ARPL, my role involves being a key contributor to the UAV Autonomy pipeline, especially in Perception, Planning and SLAM. The lab’s primary focus is on developing autonomous and agile UAVs.

Responsibilities and Accomplishments:

1. Autonomous UAV Pipeline Development in ROS2:
   • Collaborated with a team of PhD researchers to develop a UAV Autonomy pipeline in ROS2 C++.
   • Implemented quadrotor geometric controller, RVIZ-based simulation, trajectory generation module, and trajectory tracking algorithm.
   • Developed a ROS2-based VICON tracking and localizing package.
2. Modification & Deployment of OpenVINS VIO on UAV platform:
   • Customized the open-source VIO algorithm, OpenVINS, to handle platform vibrations and feature loss effectively.
   • Optimized the VIO algorithm for our specific use case on the UAV RACE platform.
   • Deployed the modified algorithm on NVIDIA Jetson boards, utilizing approximately 70% of the computation power for reliable UAV performance.
3. UAV Autonomy Pipeline for NIST UAS First Responder Challenge:
   • Engineered a customized UAV autonomy pipeline for the NIST UAS First Responder Challenge.
   • Addressed competition-specific requirements, including localization, mapping, control, GUI, and navigation.
   • Contributed to NYU’s participation in the prestigious NIST UAV First Responder Challenge, showcasing an advanced autonomous stack and achieving recognition in the competition.

At COAST Autonomous, I worked as a Robotics Engineer Intern, contributing my skills and knowledge to the development of cutting-edge technologies in the field of autonomous vehicles.

Key Responsibilities and Achievements:

1. Trajectory Prediction of Surrounding Traffic Agents: Developed a novel RNN-LSTM seq-to-seq model for predicting the motion of traffic agents in urban and high-traffic environments, achieving an impressive 80% accuracy.

2. Semantic Segmentation for Driving Space Detection: Utilized DeepLabV3 segmentation model on Cityscape dataset to accurately identify drivable spaces and obstacles in rural and construction areas.

3. Dynamic Environment Perception: Addressed challenges in AV navigation by developing solutions for navigating urban and densely populated areas.

4. Autonomous Vehicle Development: Played a role in the advancement of autonomous vehicle technology, contributing to the development of vehicles.

5. Industry Exposure: Gained practical insights into real-world autonomous vehicle projects, enhancing skills and knowledge in robotics engineering.

The career starting role at Robolab helps me to accelerate my research and product development skills in the areas of IoT, AI, and Robotics.

Key Responsibilities and Achievements:

1. Mechanical Design: Applied my experties in CAD, DFMA and CAE to create high-quality and reliable mechanical robotic systems adhering to industry-level standards..

2. Seamless Integration: Developed a comprehensive product from inception, encompassing mechanical design, electronics setup, and advanced programming for seamless integration.

3. Robotics Research Platform: Spearheaded the development of a cutting-edge Turtlebot robotic platform, seamlessly integrating a 2D LiDAR based G-mapping SLAM algorithm for mapping and localization.

4. Product Development: Designed 3 robotics-based educational kits catering to diverse age groups, fostering practical learning experiences.

5. Collaboration and Teamwork: Brainstormed and implemented innovative project ideas, designing hardware, programming software and control systems.

6. Collaborated effectively with a multidisciplinary team of engineers, educators, and designers.

7. Hands-on Workshops & Talks: Conducted engaging workshops on robotics arm and training sessions for students, teachers, and industry professionals.

More Description about project in few days once implementation is done.

Developed a Q-learning controller for a multi-variate system, and tested on torque-controlled inverted pendulum. Improved learning efficiency by 73% & reduced learning time by 15% via integration of ϵ-greedy policy.

Developed & implemented an LQR and iLQR controller for precise position & trajectory tracking of a quadrotor in disturbance-prone environments with 0.01m accuracy. It is designed with line search optimization for acrobatic maneuvers & local minima avoidance

More Description about project in few days once implementation is done.

An RNN-LSTM Seq-2Seq model which was trained on the Apollo dataset for better autonomous motion. It is able to predict future 5 positions of surrounding traffic agents based on the past 10 position sequence. It was integrated with vehicle MPC control algorithm for smooth autonomous motion.

Localize a quadrotor based on EKF and UKF by Sensor fusion of IMU and Vicon data. Developed an optical flow algorithm to localize the quadrotor and establish its velocity. Integrated visual Optical flow and IMU data to localize a quadrotor & compare with Vicon ground truth.