Feedback Design and Implementation for Integrated Posture Manipulation and Thrust Vectoring

Aniket Dhole

This MS thesis outlines my contributions to the closed loop control and system integration of two robotic platforms- 1) Aerobat, a flapping wing robot stabilized by air jets, and 2) Harpy, a bipedal robot equipped with dual thrusters. Both systems share a common theme of the integration of posture manipulation and thrust vectoring to achieve stability and controlled movement. For Aerobat, I developed the software and control architecture that enabled its first untethered flights. The control system combines flapping wing dynamics with multiple air jet stabilization to maintain roll, pitch and yaw stability. These results were published in the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). For Harpy, I implemented a closed-loop control framework that incorporates active thruster assisted frontal dynamics stabilization . My work led to preliminary untethered dynamic walking. This approach demonstrates how thrust assisted stability can enhance locomotion in legged robots which has not been explored before.

thesis (published) - December 2024

Hovering Control of Flapping Wings in Tandem with Multi-Rotors

Aniket Dhole, Bibek Gupta, Adarsh Salagame, Alireza Ramezani et al.

This work focuses on stabilizing the flight dynamics of Northeatern's tailless bat-inspired micro aerial vehicle, Aerobat. Unlike traditional insect-style designs, Aerobat incorporates morphing wings and lacks a tail, making flight control more complex. To stabilize its position and orientation during hovering, we employ a guard design with small thrusters and combine it with a flapping system and a multi-rotor. We assume the guard cannot directly observe Aerobat's states and propose the use of an observer to estimate these states for closed-loop hovering control of the Guard-Aerobat platform.

paper (published) - October 2023

Bounding Flight Control of Dynamic Morphing Wings

Aniket Dhole, Bibek Gupta, Adarsh Salagame, Alireza Ramezani et al.

Vertebrate flyers perform intermittent flights as bounding or oscillating flights for power management. Intermittent flights and the resulting oscillating height during flapping and soaring provide the means of increasing speed without increasing flapping speed. These maneuvers and their robotic biomimicry have remained unexplored so far, which, if understood, can lead to aerial robot designs with endured flight operations. This works attempts to achieve robotic bounding flight using Northeastern’s Aerobat platform. Aerobat can dynamically morph its wings by collapsing them rapidly during each gaitcycle. We present a launcher designed that allows bounding flight experimentation of Aerobat in a computer-aided fashion. We augmented Aerobat with a plural tiny thruster to stabilize its unstable roll, pitch, and yaw dynamics. This paper presents our control design based on extending Aerobat’s states to accommodate unobservable aerodynamic forces and offers experimental results to support our proposed approach.

paper (published) - August 2024

Patent - IOT BASED PORTABLE HEALTH MONITORING SYSTEM USING EDGE COMPUTING

Aniket Dhole, Mohit Gandhi, Shrishail Kumbhar, Harsh Singhal, Dr. Sonal Gore

Due to social distancing norms, several restrictions have been established in public settings due to the COVID-19 pandemic. In offices and schools, there are no automated systems or procedures for managing large groups of people. Some systems use camera footage of workspaces to verify whether individuals are wearing masks, and temperature checks are done manually by designated authorities and processed on massive servers. The paper contains a proposed prototype of a portable device that can manage if individuals entering the workspace are wearing masks, and have an appropriate heart pulse rate using M5Stack Core2, ESP32 Camera Module, and distance sensors. For optimization and fast Mask Detection Model which will run entirely on the device, Tensorflow Lite and Edge Computing are used. The mask detection model achieves an accuracy of 87.8%. Here the focus was on edge computing with limited RAM usage and with an optimized MobileNetV1 model.

patent (published) - June 2022

video

Radiomics for Parkinson’s Disease classification using Advanced Texture-based Biomarkers

Sonal Gore, Aniket Dhole, Shrishail Kumbhar, Jayant Jagtap

Parkinson's disease (PD) is one of the neurodegenerative diseases whose complete cure is not found to date. Therapies and medications are supportive methods to deal with symptoms. There is always a requirement of medical domain expertise to diagnose PD manually. Since manual diagnosis leads to a time-consuming process, an automatic technique has always been useful in such complex tasks. Magnetic resonance imaging (MRI) based computer-aided diagnosis helps medical experts to diagnose PD more precisely and fast. Texture-based radiomic analysis is carried out on 3D MRI scans of T1 weighted and resting-state modalities.Classification accuracies were obtained from 61.11% to 83.33% and area under the curve-receiver operating characteristics (AUC-ROC) values range from 0.43 to 0.86 using four variants of LBP.

paper (published) - September 2023

Review of Deep Learning Models for Mask Detection and Medical Sensors for IoT based Health Care System

Aniket Dhole, Mohit Gandhi, Shrishail Kumbhar, Harsh Singhal, Sonal Gore

The growth of medical sensors like heart rate,blood sugar, and other health monitoring sensors is huge.Along with the use of sensors in devices and healthcare systems, the use of image classification models like mask detection on edge devices is of growing demand. The survey consists of various techniques used in modern healthcare devices and various other methods like sensor fusion and wireless sensors to collect and monitor health data. And it also includes a comparison of multiple mask detection models which were deployed on embedded devices like Raspberry Pi, Nvidia Jetson and cameras like OpenMV, ESP32Cam and deep learning models like MobileNetV1, InceptionV4, and YOLO Tiny which were optimized using TensorFlow Lite.

paper (published) - Oct 2021

Parallel and Edge Computing Techniques for Computer Vision Models on Embedded Devices

Aniket Dhole, Mohit Gandhi, Shrishail Kumbhar

Nowadays, running computer vision models on embedded devices like Raspberry Pi and Nvidia Jetson has become ubiquitous. But the main issue is the limited performance on these devices due to smaller CPUs and power factors. To solve this, we have proposed research on various parallel processing techniques to get complete optimal performance of computer vision models like GoogleNet, Squeezenet, and Mobilenet on a Raspberry Pi using OpenVino Toolkit. We tested and compared these models' interpretation on factors like CPU, RAM Utilization, and Inference Time using Two Neural Compute Sticks and analyzed it on different Intel Processors. The results using Two Neural Sticks were significant than typical processors and increased by a factor of 2 to 3 for all models. So using these results, we can directly use the technique for the suitable model.

paper (accepted) - November 2021

Topical Survey on Computing Solutions for Plant Disease Classification using Deep Learning Techniques

Aniket Dhole, Mohit Gandhi, Shrishail Kumbhar, Harsh Singhal, Sonal Gore

A major problem in agriculture is plant disease that is not recognized in the early stages, due to which, the people working in this industry face resulting losses, such as lost income, loss of time and effort, etc. We have surveyed different hardware implementations of plant disease detection on embedded devices, such as the Raspberry Pi, field-programmable gate arrays (FPGAs) with very large scale integration (VLSI), and ARM processors that use frameworks such as TinyML and TFLite. And studied and analyzed major deep learning algorithms and techniques, such AlexNet, long short-term memory (LSTM), LeNet-5, and ResNet, which have been used for plant disease detection.

paper (published) - May 2021

Clarapath, New York

Robotics Engineer

Robotics Engineer Co-op

● Developed and tested an automated pathology machine leveraging ROS, precise controls, and embedded hardware for reliable operation
● Implemented automated testing and diagnostic tools to monitor encoder accuracy, sensor functionality, and overall machine performance to enable self-sustaining performance at the customer level

website

SiliconSynapse Lab, Northeastern University

Research Assistant

● Devising development of a multimodal: aerial + legged robots focusing on testing of gait optimization and contact force analysis
● Crafting position stabilization and control algorithms for aerial thrusters and leg joints, utilizing ROS, C++, Python and Matlab
● Integrated and programmed drivers for various sensors like IMU, GPS, MoCap Cameras for EtherCat network to communicate with real time Speedgoat and STM32 controllers
● Researched on Position Estimation and Autonomous control of Tiny Flapping systems in enclosed environments using SLAM and Vision based methods

website

Northeastern University

Teaching Assistant (EECE2150 : Circuits and Signals: Biomedical Applications)

● Teaching Assistant for the EECE Circuits and Signals (Biomed Apps) Course, responsible for guiding 80 students in the analysis of circuits using tools like LTSpice and Matlab.
● Collaborated with students on the implementation of intricate circuits for the purpose of reading and comprehending ECG signals, utilizing equipment such as oscilloscopes and multimeters.

website

Vuna Technologies

Co-Founder and Software Engineer

● Executed software and hardware integration for drones, cameras, and farm mapping, enabling Computer Vision-driven analysis of mango farms for plant health detection and harvest prediction using Python & Embedded C++
● Using techniques like NDVI(Normalized Difference Vegetation Index) and Object Detection to detect Disease and calculate Ripeness of fruits in farm.
● Used TinyML to deploy a Machine Learning model on a Arduino Nano BLE to predict the date of harvest using plant health and historical data
● Led and mentored a dynamic team of five developers, securing $20,000 in funding and overseeing all technical facets of startup

website

IoTIoT.in

IoT Intern

●Built a Parallel Computing Framework which automatically allocates the task to threads and cores using OpenMP.
● Parallelised Computer Vision programs like person detection, vehicle re-id using OpenCV and OpenMP to increase its speed by 220%.
● Worked on networking programs of ZMQ Server to efficiently send socket requests to server concurrently.

website

Team Automatons

Robotics Engineer and Technical Team Member

● Programmed Robotics Path Planning,Inverse Kinematics, Localization & Computer Vision on robots using Python & C++.
● Developed and analysed Embedded and Printed Circuit Boards.
● Worked on prototyping and design of Robots in mechanical manufacturing and Computer Aided Design aspects.
● Tested and implemented robot programs on Raspberry Pi, Arduino, STM32 and ESP32 Boards.

website