Trusted Polytopic Action Sets for Fast Planning in Underactuated Systems
Jaitly, A. Farzan, S
(Submitted to L-CSS)
Building autonomy by embuing hard problems with exploitable structures
Hi! I am Akshay. I am a roboticist with years of experience in research, development, and deployment for robotics problems with discrete structure, such as differentiable simulation with contact and motion planning around obstacles. My recent work has looked at GPU accelerated algorithms for differentiable collision detection, solving trust region problems, certified global optimality in nonconvex problems, and learning geometric representations of robot actions.
Also known as: Akshay, Jaitly
Publications
Papers and ResearchAnalytic Conditions for Differentiable Collision Detection in Trajectory Optimization
Jaitly, A. Jha, K D. Ota, K. Shirai, Y.
(IROS 25)
PAAMP: Polytopic Action-Set and Motion Planning for Long Horizon Dynamic Motion Planning via Mixed Integer Linear Programming
Jaitly, A. Farzan, S.
(IROS 24)
Galileo: An Efficient Pseudospectral Collocation Framework for Legged Robots
Chandler, E. Jaitly, A. Agheli, M.
(ICRA @40)
Video → · ArXiv → · Project page → · GitHub →
A MILP-Based Framework for Coordinated Multi-Agent Motion Planning and Collision Avoidance in Constrained Environments
Farzan, S. Jaitly, A. Cline J.
(CASE 25)
Experience
Aerial Control and Perception Lab. WPI - Visiting Researcher
March 2025 - Present • Worcester, MA
● Working on nonconvex optimization problems using new branch and bound tactics on the active set of constraints, solving nonlinear problems with ~6000 constraints and ~25 decision variables in ~300 ms with certificates of global optimality.
● Working on differentiable simulation of tethered payloads for vectorized simulation on GPUs.
● Designed and proved convergence of new Trust Region Problem solvers.
● Collaborating with graduate peers to integrate methods into the lab’s planning stack.
● Working on differentiable simulation of tethered payloads for vectorized simulation on GPUs.
● Designed and proved convergence of new Trust Region Problem solvers.
● Collaborating with graduate peers to integrate methods into the lab’s planning stack.
Mitsubishi Electric Research Labs (MERL) - Optimization and Intelligent Robotics
June 2024 - March 2025 & October 2025 - Present • Cambridge, MA
● Differentiable Collision Detection Massively in Parallel on GPU Utilizing Shape Smoothing (ongoing)
- Introduce a method to solve for collision/distance between objects massively in parallel utilizing GPU vectorization.
● Single Level Collision detection For Trajectory Optimization
- Outperformed other state-of-the-art methods for collision detection in an optimization program.
- Led to a publication in IROS ‘25 and a patent. See “Publications” for specific contributions.
● Learning Traffic for Elevator Scheduling
- Learning based Multi-variate time-series prediction with improved synthetic data generation.
- Dynamic-programming based algorithm to perform optimal job scheduling.
- Introduce a method to solve for collision/distance between objects massively in parallel utilizing GPU vectorization.
● Single Level Collision detection For Trajectory Optimization
- Outperformed other state-of-the-art methods for collision detection in an optimization program.
- Led to a publication in IROS ‘25 and a patent. See “Publications” for specific contributions.
● Learning Traffic for Elevator Scheduling
- Learning based Multi-variate time-series prediction with improved synthetic data generation.
- Dynamic-programming based algorithm to perform optimal job scheduling.
Boston Dynamics - Spot Manipulation Software Engineering Intern
May 2023 - August 2023 • Waltham, MA
● Contributed services for Gripper Cameras incorporated in the Spot 4.0 software release.
● Worked with image processing & legged control techniques to enhance camera calibration.
● Worked with image processing & legged control techniques to enhance camera calibration.
Autonomous Loco-Manipulation Systems Group. WPI - Research
August 2023 - May 2024 • Worcester, MA
● Worked on Galileo (see “Publications”), a software library enabling pseudo-spectral optimization for legged robot motion planning.
● This work (>60 stars) was used in other projects, including HURON (humanoid robots) and BiQu (quadrupedal robot loco-manipulation).
● This work (>60 stars) was used in other projects, including HURON (humanoid robots) and BiQu (quadrupedal robot loco-manipulation).
Featured Projects
Nonconvex optimization
Lifted homotopy continuation for global optimality in nonconvex optimization
Deriving rules for using algebraic geometry methods to solve global optimization problems tractably.
Collision Detection
GPU
Massively Parallel Differentiable Collision Detection
Using Smooth Approximations of Contact Bodies to solve differentiable collision detection on GPU.
MILP
Geometric Approx.
Polytopic Action & Motion Planning (PAAMP)
Linearizing long-horizon dynamic motion planning with learned polytopes.
C++
Galileo
Multi-Contact Planning on Unitree Go1
Perceptive locomotion: walking, hopping, trotting with switched-system collocation.
C++
Arduino
LLAMA-Q: A C++ Library to Abstract and Generalize Robot Control
MIT THINK award finalist; granted a presentation slot at MakerFaire 2020.
Differentiable Simulation
Differentiable Simulation of Tethered Payloads for Trajectory Planning
In progress...
Matlab
Geometry
Degen_vert2lcon: A MATLAB library to find the Convex Hull of points that lie in an affine subspace
Published on MATLAB file exchange, this addressed limitations of existing methods for finding convex hulls.
ROS
TrajOpt
Trajectory Tracking for Quadrotor
Robust Trajectory Tracking for Quadrotor using Sliding Mode Control.
Swarm Communication
Distributed Task Allocation for Communication in Intermittent Swarms
Solving a distributed optimization with analytic bounds on communication frequency.
Kinematics
Optimization
Wirebot Platform
A < $50 platform for speedy locomotion in large indoor spaces.
Kinematics
TrajOpt
Hephaestus Arm Control
Implementing and using Vision, Position/Velocity Kinematics, Trajectory Planning, Communication to enable robot control.
Teaching
Grades 7–12 • Semester-long courses
● Invited to develop curricula and teach courses in Linear Algebra, applied math, and Robotics for students in grades 7–12.
● Oversaw completion of student projects, including satellite localization algorithms and prosthetics development.
● Oversaw completion of student projects, including satellite localization algorithms and prosthetics development.
Curriculum Builder
● Built curriculum on kinematics, microcontrollers, and related topics for Curious Cardinals, a startup out of Stanford.
Education Service • 85k+ students
● Directed new projects for STEMpump, a student-led education service with over 85k students worldwide, overseeing pedagogy and new course development.
Global Robotics Education
● Collaborated with ESL-focused teachers to develop hands-on robotics curriculum for the affordable XRP platform.
● The XRP project has been used to teach robotics worldwide.
● The XRP project has been used to teach robotics worldwide.
About
Hi! I am Akshay. I am a roboticist with years of experience in research, development, and deployment for robotics problems with discrete structure, such as differentiable simulation with contact and motion planning around obstacles. My recent work has looked at GPU accelerated algorithms for differentiable collision detection, solving trust region problems, certified global optimality in nonconvex problems, and learning geometric representations of robot actions.
Interests: Optimization, Algebraic Geometry, Motion Planning, Switched Systems, Underactuated Control.