Akshay Jaitly - Resume 1

Akshay Jaitly 55 Institute Road, Worcester,
Mass. 01609-2280
Worcester Polytechnic Institute - 2024
(469) 968-7924
Bachelors and Masters in Robotics Engineering akshayjaitly@hotmail.com
akshay-jaitly.weebly.com
Tools: C++, Python, MATLAB, Eigen, Drake, CasADI, LaTeX, Git, Bash, Arduino, ROS, Gazebo, Fusion 360
Interests: Controls and Motion Planning, Athletic Intelligence, Optimization Methods (Nonlinear and Mixed Integer)
Work & Lab Experience
Boston Dynamics - Spot Manipulation Software Engineering Intern (May 2023 - Aug 2023.)
➢ Contributed to Boston Dynamics’ extensive perception and controls code base to develop new services.
➢ Part of my work, which is implemented in Spot 4.0, involved developing effective hand-eye camera calibration.
➢ Designed experiments with real robots to find errors and improve accuracy, and prove efficacy of approaches.
➢ Worked with image/data processing (object detection, PnP, and Least Squares methods) and legged control.
Autonomous Loco-Manipulation and Systems (ALMaS) Group @ WPI (Aug 2023 - pres.)
➢ Research and work developing frameworks for Loco-Manipulation and planning through contact.
➢ Development of Galileo, an open source library for Pseudospectral Collocation on switched systems
➢ Galileo is being put to work on HURON/HERO (ALMaS’ humanoid robot) and BiQu (Quadrupedal robot project)
➢ Working with teams to integrate visual SLAM and manipulation on these legged robots.
SE Watson and Associates — Product Development (Aug 2022 - Dec 2022.)
➢ (Under NDA) Facilitated Software and Embedded Systems development for a medical device. This included
prototyping hardware and developing software.
Preprints, Publications & Submissions
1. Jaitly, A., & Farzan, S, “PAAMP: Polytopic Action-Set and Motion Planning for long horizon dynamic
motion planning via Mixed Integer Linear Programming”. Submitted to IROS 2024.
https://arxiv.org/abs/2403.10924 -- PAAMP on ArXiv
● Uses learned polytopic representations of behaviors to simplify nonlinear dynamic motion planning.
● Approximating robots as switched systems inherently led to a fast Mixed Integer Linear Program that
could be used to solve for feasible longer horizon motions in less than a millisecond.
● The use of these representations of behaviors enables solving for motions that naturally follow robot
dynamics, rather than attempt to counteract them.
● Heuristics are created to efficiently guide exploration of solutions to the MILP inspired by domain
analysis in Task And Motion Planning / Multi-Modal Motion Planning,
● This leads to applications for TAMP, dexterous robots and legged robots with comparable speeds.
2. Chandler. E., Jaitly, A., & Agheli, M, “Galileo: An efficient Pseudospectral Collocation Framework for
Legged Robots”. Submitted to IROS 2024.
● Galileo is an open-source library implementing pseudospectral collocation to solve for robot motion.
By transcribing deviants from the state as decision variables, we were able to efficiently optimize walking
motions for quadrupedal robots and plan through contact.
● Results show trotting, hopping, and walking on the Unitree Go1, Huron, and Boston Dynamics’ Atlas.
● Collocation methods like this implicitly solve for dynamically feasible trajectories, unlike methods reliant
on Runge-Kutta or Euler methods.
Projects
Learning Polytopes of Dynamic Feasible Trajectories in Planning Space
➢ Adapted IRIS-NP to learn more natural and computationally cheap representations of behaviors for PAAMP.
➢ IRIS-NP becomes infeasible in abstract spaces that consider complex robot dynamics. Methods to generate more
appropriate data points to learn dynamic spaces needed to be developed.
➢ Implemented with Drake for model based creation and verification of behavior “polytopes”.
Tools: Drake, Eigen, MATLAB Topics: Behaviors, Planning, NLP, SDP, Computational Geometry
BiQu: Legged Perceptive Locomotion (team)
➢ Creating a ROS implementation for perception and control of a real Unitree Go1 quadruped.
➢ Led the control implementation, built using Galileo, our open source library for collocation in switched systems.
➢ We can ensure motion over long periods of time by extending Galileo as a Model Predictive Controller.
➢ Contributed heavily to negotiations on the Unitree Robot, leading to a ~75% discount.
Tools: ROS, CasADI, Eigen Topics: SLAM & Perception, Whole Body Control, Motion Planning
Task Allocation for Data Mules in Intermittent Networks of Robots (team)
➢ Led the development of a distributed solver for task allocation in swarms of robots with intermittent
communication. The solver did not rely on consensus (an expensive task) to decide on “data mules”.
Tools: ARGoS Topics: Swarms, Networks, Distributed Optimization
2
Creating a Robust Sliding-Mode Controller for a Quadcopter in ROS (team)

➢ Implemented a sliding-mode controller for control of the CrazyFlie 2.0 quadcopter.
➢ Experiment with usage of activation functions to maintain robustness while applying less control effort.
Tools: MATLAB, Python, ROS, Gazebo Topics: Feedback Control, Aerial Vehicles
BasicLinearAlgebra (contributor)
➢ Contributor on BLA, a top open source library (>175 stars) for Linear Algebra on microcontrollers
➢ Contributed Vector Valued Functions, computation of the Jacobian, and a “Ones” vector, used for LLAMA-Q
➢ Designed tests to validate the continued validity of contributions made.
Tools: Arduino Topics: Linear Algebra, Open Source
LLAMA-Q: a C++ Library and Approach to Abstract and Generalize Robot Control
➢ I created LLAMA-Q to enable high level experimentation with robots on Arduino for beginners and hobbyists.
➢ Wrote Gradient Descent and SQP optimizers from scratch to develop a general kinematics solver, and
successfully applied them to a “wirebot”, a cable-actuated mechanism with the potential to scale large buildings
➢ This project was an MIT THINK Finalist, and was presented at Maker Faire 2020.
Tools: Arduino, Fusion 360 Topics: Algorithms, Data Structures, SQP, Embedded Systems
Navigation and Mapping on Wheeled Robots (team)
➢ Created a ROS package for navigation of a wheeled robot with an integrated LIDAR.
➢ Navigation consisted of simple grid based path planners (RRT, A*).
➢ Implemented both Kalman filters and Particle Filters (at different points in time) for localization.
Tools: ROS Topics: SLAM, Particle Filter, Sensor Fusion, Navigation
Volunteer/Work in Pedagogy
Mentor, Curious Cardinals (Dec 2020 - July 2022)
➢ Advised middle/high school robotics projects and built curriculum on kinematics, microcontrollers, etc.
➢ Worked with classrooms around the country to develop and teach new courses.
STEMpump (Aug 2021)
➢ Directed new projects for STEMpump, a student-led education service with over 85k students worldwide. This
included overseeing legal incorporation, fundraising, and new course development.
Curriculum Development for the Experiential Robotics Platform ( Jan 2023 - May 2023 )
➢ Worked with teachers to develop and improve curriculum for an affordable hands-on robotics platform (XRP)
➢ This curriculum has been used internationally, being presented at the White House and the UK PM’s office.

Akshay Jaitly - Resume 1

Uploaded by

Copyright:

Available Formats

Akshay Jaitly - Resume 1

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Akshay Jaitly - Resume 1

Uploaded by

Copyright:

Available Formats

Akshay Jaitly 55 Institute Road, Worcester,

Creating a Robust Sliding-Mode Controller for a Quadcopter in ROS (team)

You might also like