This work is a continuation of the work I did while I visited Robotics Research Centre at Nanyang Technological University, Singapore as a Summer Research Student from May to July 2024.

Abstract:

The non-commutative nature of 3D rotations poses well-known challenges in generalizing planar problems to three-dimensional ones, even more so in contact-rich tasks where haptic information (i.e., forces/torques) is involved. In this sense, not all learning-based algorithms that are currently available generalize to 3D orientation estimation. Non-linear filters defined on SO(3) are widely used with inertial measurement sensors; however, none of them have been used with haptic measurements. This paper presents a unique complementary filtering framework that interprets the geometric shape of objects in the form of superquadrics, exploits the symmetry of SO(3), and uses force and vision sensors as measurements to provide an estimate of orientation. The framework’s robustness and almost global stability are substantiated by a set of experiments on a dual-arm robotic setup.

Course: Advanced Process Control, Prof. Mani Bhushan  

• Modeled the 2-DOF robotic arm as a non-linear dynamical system. 

• Performed open loop simulation to determine characteristics of the system such as settling time, etc. 

• Obtained discrete-time linear perturbation model, implemented Model Predictive, Linear Quadratic Optimal & Pole Placement controllers, and compared performance for regulatory & servo problems in MATLAB. 

Course: Control of Nonlinear Dynamical Systems, Prof. Srikant Sukumar 

Designed & simulated a (i) backstepping and (ii) passivity-based torque control law for the manipulator to track reference constant using MATLAB. 

Course: Linear Systems Theory, Prof. Vivek Sangwan 

Surveyed the use case of Singular Value Decomposition (SVD) to determine the singularities of a manipulator. 

In this project, a UR-5 robotic arm and a mobile robot collaborate to sort and prepare packages autonomously for shipment from a life-size warehouse at IIT. I worked on- 

• Building the hardware of the entire system, including the upgradation of the mobile rover's embedded computing device from NVIDIA Jetson TX2 to NVIDIA AGX Xavier and then finally to Intel NUC, and porting and creating hardware codes for ROS2.   

• Creating software packages for remote operation of the entire system using ROS2.

• Developing safety scripts for safe operations of the robots especially when controlled remotely.

• Localization of the mobile rover using Simultaneous Localization and Mapping (SLAM) toolbox in ROS2. 

• Developing the simulation environment on Gazebo and optimizing it for better performance on low-end laptops.

• 146 students learned to create autonomous algorithms with 30.8 hours of average access to hardware. 

1,312 students were trained using this system on remote hardware and simulator as part of the e-Yantra Robotics Competition. This work recently got accepted at the 16th International Conference on Social Robotics, Odense, Denmark with me as the lead author.

In this project, a UR-5 robotic arm mounted on a mobile robot work together to identify and pluck bell peppers autonomously from a life-size greenhouse at IIT. I worked on- 

• Creating a full software stack for controlling the robots remotely using a peer-to-peer VPN.

• Improving, maintaining, and updating the hardware of the mobile rover, including debugging and solving issues related to the power supply for the NVIDIA Jetson TX2 board, the addition of a heat management system, especially for the high temperatures (~122°F) inside the greenhouse, etc. 

• Developing safety scripts for safe operations of the robots especially when controlled remotely. 

• Creating software packages for remote operation of the entire system using ROS.

• Developing the simulation environment on Gazebo and optimizing it for better performance on low-end laptops.

• 70 students learned to create autonomous algorithms with 19.4 hours of average access to hardware. 

1,433 students were trained using this system on remote hardware and simulator as part of the e-Yantra Robotics Competition. This work recently got accepted at the 16th International Conference on Social Robotics, Odense, Denmark, with me as the lead author.

In this project, a nano drone is localized using a single monocular camera mounted on the ceiling. A WhyCon marker is added on top of the drone. Once localized, the arena is emulated in V-REP simulator to plan paths using OMPL. ROS is used to integrate the drone commands with the V-REP simulator. With a localization error of 3.1cm and a setup cost of 50 USD, this approach is ideal for environments where cost is a significant concern. I worked on-

• Developing an Iterative Feedback Auto-tuning algorithm for the PID controllers using ROS.

• Emulating the real-world arena in the V-REP simulator, including finding transformation for both the coordinate frames.

• Finding paths using OMPL and commanding the drone even when feedback from the WhyCon was temporarily unavailable.

This project was part of my summer internship at IIT Bombay and the e-Yantra Robotics Competition in 2018. Along with other applications developed using this platform, we have submitted a paper to Cyber-Physical Systems Journal and is currently Under Review. 

In this project, I used Simscape to model a flexible quadruped.

Different colors represent different links. Link red-green has a torsional spring in between. Link red-pink and green-pink have a revolute joint. Links pink-black have a linear spring. Links yellow-blue represent revolute joints in between. Black links are broader for stability.

Contact forces are simulated by exporting the geometry of the legs and the ground. Parameters, including friction, are shown in the table above.

The front motors (revolute joints) are provided with a sinusoidal input, while the back ones are provided the same input with an offset such that these inputs are out of phase.