Robotics Engineer
M.S. | Mechanical Engineering | University of California San Diego | USA
B. Tech | Mechanical Engineering | IIT Bombay | India
Research Interests:
Modeling, Simulation, and Control of Robots, Path planning,
Robot Navigation, Manipulation and Locomotion
Email: a4singhal@ucsd.edu
LinkedIn | Github
SCHOLARLY ARTICLES
Animesh Singhal, Sahil Modi, Abhishek Gupta, Leena Vachhani, Omkar Ghag, “Pendulum Actuated Spherical Robot: Dynamic Modeling & Analysis for Wobble & Precession” accepted to the 22nd IFAC International Symposium on Automatic Control in Aerospace (ACA) 2022. | [Preprint available] [arXiv]
(Received the award for Best Presentation at the ACA conference on Nov 24, 2022)
Animesh Singhal, Sahil Modi, Abhishek Gupta, Leena Vachhani, “Wobble control of a pendulum actuated spherical robot”. (Submitted to Robotica, a Cambridge University Press Journal) | [Preprint available] [arXiv]
Leena Vachhani, Abhishek Gupta, Vaibhav Nandkumar Kadam, Kaustubh Prashant Sadekar, Animesh Singhal, “A Spherical Robot”, Application No.: 202223063422, filed on 7th November, 2022
Leena Vachhani, Abhishek Gupta, Vaibhav Nandkumar Kadam, Kaustubh Prashant Sadekar, Animesh Singhal, “Methods for scaling a spherical robot”, Application No.: 202221063295, filed on 5th November, 2022
Leena Vachhani, Abhishek Gupta, Kaustubh Prashant Sadekar, Vaibhav Nandkumar Kadam, Animesh Singhal, “A robot system with an upwards-facing camera”, Application No.: 202221067357, filed on 23rd November, 2022
WORK EXPERIENCE IN ROBOTICS
During my 6-month tenure at Black Coffee Robotics (BCR), my focus encompassed Human-Robot Interaction (HRI), Robot Path Planning, and Robotic Simulator Analysis.
To foster HRI, I designed a Gazebo plugin for simulating human behavior and created an AMR navigation testbed, showcasing realistic human interaction scenarios. I also conducted a comprehensive comparison of popular global and local path planning algorithms to optimize path generation for a differential drive robot. My findings revealed that SMAC outperformed SBPL, offering a threefold speed advantage in producing kinematically feasible paths. Additionally, I devised Gazebo-based test environments to assess various local planners, such as DWA and TEB, under diverse conditions, spanning different speeds and confined spaces. I also assessed the performance of the recently open-sourced robotic simulator, Webots, by subjecting it to SLAM toolbox testing on an AMR.
My time at BCR proved instrumental in refining my skills in ROS, ROS2, Python, C++, Git, Docker, and general software development best practices. It was an invaluable experience, offering insights into the dynamics of a deep-tech startup. I gained a wealth of knowledge about project management and teamwork, even while operating in a remote work setting.
Problem:
Our in-house pendulum-actuated spherical robot at IIT Bombay’s Embedded Control Lab has a coupling between steering and linear motion. Due to this, the bot operates with lateral oscillations at low speeds. As a result, the surveillance camera mounted on the robot provides shaky feedback. To circumvent this, we sought to control the under-actuated robot with only two inputs to move along simple trajectories, such as a circle or a straight line, without wobbling. Another constraint is that the pendulum must move in a stable manner in our robot’s lower hemisphere, as the upper hemisphere houses sensors and other electronics.
Approach:
Modelled the wobble present in the motion of a spherical robot by developing equations of motion that account for the coupling of its forward and steering motions.
Developed formulations for the wobble amplitude and frequency, turning radius, and the precession rate associated with the robot's circular motion to demonstrate their relationship with robot’s speed and pendulum angle.
Designed a model based controller design for controlling a pendulum-actuated spherical robot's heading direction via a turning maneuver while limiting its wobble and maintaining a stable and constrained pendulum motion. The control set-points are determined by calculating the desired robot speed and pendulum angle based on the required turning radius of the turning maneuver.
Result:
The approach mentioned above would be applied to operate the robot semi-autonomously. In this mode, the operator would command a specific pendulum angle to steer the robot in the desired direction while maintaining a specified forward speed. The designed controller would then assist the robot in moving without wobbling along the desired circular arc or line.
ROBOTICS PROJECTS
Designed an input-output feedback linearization based controller to control spherical robot’s pendulum, operational speed, and lateral oscillations
Executed cascaded PID controller for drone with speed & throttle constraints; capable of handling model-errors
Implemented PID controller to keep car on track by appropriately adjusting steering angle & parameter tuning
Developed an open--source Gazebo plugin to dynamically control simulated humans through velocity/path commands sent via ROS topics.
[Media credits: Black Coffee Robotics]
Modelled an underactuated pendulum-actuated spherical robot using Lagrange-D'Alembert equations while taking the non-holonomic constraints into account
Derived a dynamic quadrotor model, considering propeller-generated forces and moments for acceleration and rotation predictions
Generated motion primitives to create kinematically feasible paths for free/cart-joint bots for graph search methods. Evaluated and benchmarked lattice-based planners viz. SMAC & SBPL to generate geometrically predictable paths.
[Media credits: Black Coffee Robotics]
Built Path Planner to generate waypoints on virtual highway with traffic; created collision free & smooth trajectories
Implemented Kalman Filter on simulated Lidar & Radar measurements to track a bicycle around a self-driving car
Applied 2-D Particle Filter to localize self-driving car using GPS, Lidar & Radar data; achieved cm-level accuracy
Integrated Webots simulator & ROS2 to test SLAM-toolbox on AMRs. Reported flaw in LiDAR's measurement update to maintainers.
[Media credits: Black Coffee Robotics]
Built lane-finding algorithm using distortion correction, sliding window, color transforms, & gradient thresholding
Deployed CNN for driving in a simulator by cloning human driving behavior to predict steering angle using Keras
Programmed ROS nodes to control car’s acceleration & steering to navigate simulator's waypoints & respond to traffic lights
Created Gazebo-based test-bed for AMR navigation amidst realistic human crowd simulated using Social Force Model
[Media credits: Black Coffee Robotics]
