Research

Predictable multi-core implementation of multi-rate sensor fusion for high-precision positioning systems

  • A predictable and composable multi-core system-on-chip (MPSoC) platform designed for developing embedded control applications with strict performance requirements.
  • Predictable multi-core implementation of multi-sensor fusion imposing (nearly) constant delay.
  • Reducing sensing delay using predictable parallel and pipelined implementation.

Improved Positioning Precision using a Multi-rate Multi-sensor in Industrial Motion Control Systems

  • A multi-rate multi-sensor fusion algorithm fusing accurate but slow and delayed vision sensor data and fast but less accurate linear encoder data along with a bias correction solution to correct the effect of external disturbances.
  • The proposed fusion algorithm offers a higher-accuracy positioning estimation at a high operating frequency.
  • An evaluation framework to analyze and optimize the performance of multi-rate multi-sensor industrial motion control systems.

An Evaluation Framework for Vision-in-the-Loop Motion Control Systems

  • The SIL simulation to evaluate the effects of vision sensor feedback and the operating environment on the motion control system.
  • The PIL simulation of C code for discrete-time controller and image processing algorithm on an instance of the CompSOC platform. This feature further establishes interfacing with CoppeliaSim using application programming interfaces (APIs).
  • Automatic code generation for the VIL system software targeting the CompSOC platform.

FPGA Implementation Framework for Accelerating Nonlinear MPC Through Machine Learning

  • Proposed a framework for accelerating Nonlinear Model Predictive Control(MPC) through Machine Learning
  • Approximated control law using Deep Neural Network(DNN) for a nonlinear system
  • DNN-based NMPC implemented on a Xilinx’s ZYNQ-7000 SoC ZC706 FPGA board using low-level C/C++ code and demonstrated for a flying robot control application.

A Memory Efficient Implementation of Offset-Free Explicit Model Predictive Controllers on FPGA using Posits

  • Proposed a novel approach for memory efficient embedded implementation of MPC using Posit numbers
  • HIL co-simulation for Posit-based offset-free Explicit MPC on FPGA
  • Detailed analysis of posit based offset-free EMPC such as FPGA resources utilization, power utilization, Posit operations per second
  • Demonstrated on Citation Aircraft and Anesthesia control model

A Framework for High accuracy low precision embedded model predictive control using posits

  • Proposed a high accuracy and low memory quadratic programming solver using posit numbers
  • Proposed Posit-based QP solvers
  • Embedded implementation framework for posit based MPC
  • Demonstrated on Couple tank, Double integrator, Benchmark convex QP problem repository

A Memory efficient explicit model predictive control using posits

  • Proposed a novel approach for memory reduction in Explicit MPC
  • Posits based EMPC
  • Memory utilization calculation and comparisons with classical IEEE-754 number based MPC.
  • Couple Tank Control

Development of Posit Arithmetic and Logical Unit (ALU) in Verilog and its HIL Co-simulation on FPGA

  • Developed Verilog functions for posit addition and multiplication
  • Tested with Xilinx system generator in MATLAB Simulink
  • HIL co-simulations of all posit arithmetic on FPGA

Stochastic MPC using Stochastic Gradient Descent (SGD), Stochastic Proximal Point (SPP)

  • Development of SGD, SPP, proximal point algorithms in MATLAB
  • Convergence rate analysis for large scale standard datasets
  • Non-smooth optimization, composite optimzation
  • Solving optimization problems such as l2 SVM, Hinge loss, logarithmic regression

Design and development of first order unconstrained and Constrained optimization methods

  • Development of Steepest descent, Newton Method, Conjugate gradient and BFGS, Penalty barrier, SQP optimization methods
  • Performance comparison for checking convergence rate and simulation time
  • Embedded implementation for improving timing constraints

Linear and Explicit MPC on FPGA

  • Implemented Interior Point Method, Active Set Method for Linear MPC.
  • MPT toolbox, Hybrid toolbox
  • Framework for embedded implementation using FPGA, STM Microcontrollers
  • Couple tank, Anesthesia, Aircraft, 2DOF helicopter control

Framewok design in MATLAB for auto tuning of PID Controller

  • Developed Ziegler-Nichols tuning method in MATLAB for Auto tuning of PID controller
  • Verification of Kp, Ki and Kd parameters values using inbuilt MATLAB auto tune features to compare the performance.
  • C code development for embedded implementation
  • Designed GUI for PID auto-tuner for Type-I,II and III transfer functions
  • GUI developed with features of closed loop response, PID tuning parameter, C code generation

Mathematical modeling and simulations

  • Two tank, Quadruple Tank System
  • Anesthesia control problem
  • Citation Aircraft
  • 2DOF helicopter
  • Inverted Pendulum, Crane model
  • Boiler, Heat exchanger
  • DC motor