Shambhavi Singh

I am a Research Staff with Prof. Howie Choset at the Biorobotics Lab, Robotics Institute , part of the School of Computer Science at Carnegie Mellon University . Over the past year, I've dived deep in multi-agent task planning and path finding problems using optimization-based methods and hierachical planning methods. Next, I am working on a project involving multi-agent coverage and exploration.

I have a Masters of Science in Physics and a Bachelors of Engineering in Electrical and Electronics from BITS Pilani where I worked with Prof. Ashwin KP on parallel manipulator systems, Prof. Sarang Dhongdi on ad-hoc aerial networks for search and rescue operations, and Prof. Rakesh R. Warier on distributed coordination of networked multi-agent systems.

I am applying for PhD positions starting Fall 2024, with a focus on planning for multi-agent systems.

Email  /  Github  /  Google Scholar  /  Linkedin  /  CV

photo of Shambhavi smiling

Research

Animation of structure being built

Hierarchical Planning for Long Horizon Multi-agent Collective Construction


Shambhavi Singh, Zejian Huang, Akshaya Srinivasan, Geordan Gutow, Bhaskar Vundurthy and Howie Choset
accepted at International Conference of Robotics and Automation (ICRA), Tokyo, Japan, 2024
website /

We present a hierarchical approach that first finds a sequence of tasks to complete a goal, then identifies ordering constraints between tasks to parallelize for multiple agents, and finally computes collision-free paths for agents.

Pipeline showing four steps of hierarchical planner

Hierarchical Propositional Logic Planning for Multi-Agent Collective Construction


Shambhavi Singh, Akshaya Srinivasan, Geordan Gutow, Bhaskar Vundurthy and Howie Choset
Workshop on Future of Constrution, International Conference on Robotics and Automation, London, UK, 2023
paper /

We task teams of robots to collectively construct a modular structure composed of blocks. We employ temporal logic to plan at the task-level (robots pick-up and place blocks), then parallelize and find paths for multiple agents (robots move while carrying blocks) to execute the tasks.

A structure decomposed in 5 parts

Multi-agent Collective Construction using 3D Decomposition


Akshaya Srinivasan, Shambhavi Singh, Geordan Gutow, Howie Choset and Bhaskar Vundurthy
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Detroit, USA, 2023
arxiv / website / paper /

We use Mixed Integer Linear Programming (MILP) to find a multi-robot plan to construct a modular structure composed of cubic blocks. We use a 3D decomposition based method to first find plans that construct smaller substructures, then aggregate the solutions to obtain a complete plan.





Other Projects

These include coursework, side projects and unpublished research work.

Modular robot moving on flat ground

Latent Space Optimization for Modular Robot Design


Research Project - Carnegie Mellon University
2023

Designing a VAE to structurally encode robot designs (in known environments) such that we can use Bayesian Optimization to converge on well performing robot designs for unseen environments.

Gif of parallel robot platform and voltage source in background

Development and control of 3RPS Parallel Manipulator System


Mechanical Dept Research Project - BITS Pilani
2021
code /

Research Project with Dr. Ashwin KP at BITS Pilani, Goa. Modeled the dynamics and control of a 3RPS Parallel Manipulator platform. Developed the robot with three linear actuators and an acrylic base.

Graphic of a drone near a forest fire

Analysis and Development of Ad-hoc Drone Networks for Disaster Management


Electrical and Electronics Dept Research Project - BITS Pilani
2021

Research Project with Dr. Sarang Dhongdi at BITS Pilani, Goa. We built a framework for the analysis of ad-hoc networks of aerial vehicles under constrained mobility conditions. This work was developed for robust and reliable exploration of disaster prone areas by teams of aerial vehicles. The framework used NS3 and ROS, and we further emulated the network on Intel boards that can be mounted on real drones.

Custom built drone used for autonomous trajectory tracking

Quadcopter Trajectory Tracking


Aerodynamics Club, BITS Pilani
2020

Autonomous trajectory tracking with a custom-built drone using PID control. Implemented using ROS and PX4 Autopilot for processing odometry data for position estimation and tracking

Multiple swarms of birds flying in the sky

A Survey on Efficiency of Multi-Swarm Systems


Embedded System Design Course Project - BITS Pilani
2020
paper /

For a course project on Embedded Systems Design, we review algorithms used in Multi-Swarm Systems, with a focus on Particle Swarm Optimization techniques. In particular, we present a detailed analysis of algorithms for task allocation, communication and grouping for multi-swarm systems.

Mars Rover Robot with gripper equipped manipulator system in 2019

6 D-O-F Manipulator Arm for Mars Rover


Mars Rover Team (Kratos) - BITS Pilani
2020
code /

Development and inverse kinematic control of end-effector for heavy-object manipulation (carrying loads), and precision manipulation tasks (soil sample testing). Also implemented Model Predictive Control of end effector position.

Four black and white modules attached in pairs

Self-Reconfigurable Modular Robots


Robotics Club - BITS Pilani
2019

Led mechanical design team to design a novel and secure module-locking mechanism in Solidworks for self-reconfiguration of identical modules. Manufactured using Ultimaker 3D Printer.

Plot of a Bi-layer Graphene with Yellow Carbon Molecules

Density Function Theory Experiments for materials in Semiconductor Devices


Physics Dept Research Project - BITS Pilani
2019
code / report /

Simulating and visualizing properties of Metals (Al, Cu, Ni, Fe and Ag) and Single and Bi-layer Graphene, implemented in Fortran.

Graphical representation of a Processor or CPU

Hybrid Round-Robin Process Scheduling Algorithm


Operating Systems Course Project - BITS Pilani
2019
code /

Development of scheduler optimized for long- duration processes and high-traffic loads using Shortest Job First with Round-Robin and dynamic quantum times.


Design and source code from Jon Barron's website