This work presents a framework for self-supervised one-shot segmentation of synthetic images generated using StyleGANs. The framework consists of a SwAV-based contrastive learner that uses GAN style features to perform on-the-fly one-shot segmentation of GAN-generated images.

Paper:
Manerikar, Ankit, and Avinash C. Kak. Self-Supervised One-Shot Learning for for Automatic Segmentation of StyleGAN Images. arXiv preprint arXiv:2303.05639 (2023).

Code: GANecdotes Repo

Video: Model Demo

A major problem in designing ATR classifiers is the difficulty in generating and obtaining X-ray datasets containing hazardous materials and objects. An efficient alternative is to simulate this X-ray data for training and testing the ATRs. This work consists of an ML-based model that uses a StyleGAN2-ADA model for synthetising baggage X-rays scans for prohibited item detection .

Code: BagGAN Repo.

This work proposes an alternate physics-based implementation of an automated dataset simulation pipeline for ATRs for different types of threat objects. The simulator captures the dynamics of baggage configurations, concealment of threat items within baggage and the physics of X-ray scanning and image generation

Paper:
Manerikar, Ankit, Fangda Li, and Avinash C. Kak. DEBISim: A Simulation Pipeline for Dual Energy CT-based Baggage Inspection Systems. Journal of X-ray Science and Technology 29.2 (2021): 259-285.

Code: DEBISim Webpage, GitHub

Automatic Target Recognition involves designing classifiers for automatic delineation of threat regions from checked baggage scans. But given the changing security scenarios, these classifiers have to deal with varying target specifications for carrying out the segmentation. In this research, we present a two-stage classifier structure that achieves this without re-configuring the entire EDS.

Paper:
Manerikar, Ankit, Tanmay Prakash, and Avinash C. Kak. Adaptive target recognition: A case study involving airport baggage screening." Anomaly Detection and Imaging with X-Rays (ADIX) V. Vol. 11404. International Society for Optics and Photonics, 2020.

Dual Energy Computed Tomography deals with scanning a baggage volume with two different X-ray spectral sources simultaneously. This allows extracting additional material-based information from the collected projection data such as the effective atomic number. This research deals with processing dual energy data for better identification of threat materials.

Papers:

Manerikar, Ankit, Fangda Li, and Avinash Kak. "A Spectrum- Adaptive Decomposition Method for Effective Atomic Number Estimation using Dual Energy CT." IS&T Electronic Imaging: Computational Imaging VIII, IS&T International Symposium on Electronic Imaging, 2020.
Li, Fangda, Ankit Manerikar, Tanmay Prakash, and Avinash Kak. "A Splitting-Based Iterative Algorithm for GPU-Accelerated Statistical Dual-Energy X-Ray CT Reconstruction." IS&T Electronic Imaging: Computational Imaging VIII, IS&T International Symposium on Electronic Imaging, 2020.

Surveying and inspection applications for mobile robotics present a unique problem for robot path planning. Unlike a normal path planning task which involves figuring a path from a given source and destination, surveying robots require coverage path planning, i.e., calculating a path that maximizes the area covered by the robot from a given area. We have developed several implementations of coverage path planning for different applications involving indoor and outdoor mobile mapping platforms (MMS) for photogrammetry applications.

Papers:
Shamseldin, Tamer, Ankit Manerikar, Magdy Elbahnasawy, and Ayman Habib. "SLAM- based Pseudo-GNSS/INS localization system for indoor LiDAR mobile mapping systems." In 2018 IEEE/ION Position, Location and Navigation Symposium (PLANS), pp. 197-208. IEEE, 2018.
Manerikar, Ankit, Tamer Shamseldin, and Ayman Habib. "SLAM-Assisted Coverage Path Planning for Indoor LiDAR Mapping Systems." arXiv preprint arXiv:1811.04825 (2018).
Manerikar, Ankit, Debasmit Das, and Pranay Banerjee. "Optimal Control for Constrained Coverage Path Planning." arXiv preprint arXiv:1708.03055 (2017).

The RMPD (Recursive Mid-Point Displacement) algorithm is a novel way of calculating robot paths and trajectories developing for 6-DOF robots for automatically pruning apple trees branches while collisions.

Papers:
Li, Fangda, Ankit V. Manerikar, and Avinash C. Kak. "RMPD—A Recursive Mid-Point Displacement Algorithm for Path Planning." In Twenty-Eighth International Conference on Automated Planning and Scheduling. 2018.

The topic deals with the problem of visual place categorization (VPC), i.e., the visual identification of an input scene as belonging to a particular semantic pre-trained category. The place detector is implemented with a two-tier structure of a CNN trained for place recognition and a clusterer for partitioning maps for different locations.

Code: Indoor Place Categorization for Visual SLAM
Slides: YouTube

The project involved the implementation of an accoustic levitation chamber which allows precise suspension and levitation of particles in mid-air. The levitation phenomenon is created by generated resonant accoustic pressure waves from an ultrasonic horn and by maintaining the resonance by precision control of the cavity length.

More Info: pdf

It involves the study of tripedal walking mechanism with only one active limb in the center. The combination is still able to mimic bipedal motion with an interesting combination of active and passive dynamic walking mechanisms.

More Info: GitHub

Here we have implemented a free-falling mechanism that mimics the cat-righting reflex - the ability of a cat to land on all its four legs during a free-fall. The mechanism uses controlled actuation to mimic the non-holonomic centripretal forces that cause the righting reflex.

Paper: Manerikar, Ankit, and Anandpara, Tanvi. "Design of a Practical Cat-righting Reflex (CRR) Model." Procedia Computer Science 45 (2015): 514-523.