Ashvin N Iyer

I'm a 4th year undergraduate at Purdue University, majoring Computer Science and minoring in Mathematics. I love anything computer science related, but I'm especially interested in Robotics and Artificial Intelligence.

Most recently, I interned at Kodiak Robotics, working on simulation and motion planning for self-driving trucks (and will be returning for full-time in January 2025!) I have also spent time performing robotics research at both the IDEAS and CoRAL Labs, where I have explored both classical and learning-based methods for robotics.

Email / CV / Github / Linkedin

Experience

	Simulation / Motion Planning Software Engineer Intern - Kodiak Robotics September 2023 - December 2023 / May 2024 - August 2024 I worked with both simulation and motion planning teams to help improve the performance of our planning algorithms. Specifically, I enhanced simulator capabilities to support testing of the entire planning stack, automated large-scale evaluation of generated plans in simulations, and improved upon existing plans using sampling-based techniques.
	Embedded Software Engineer Intern - Amazon May 2023 - August 2023 I worked on the bluetooth team in Lab126, where I helped enable simultaneous bluetooth device connections to single audio source and handle multi-client phone call logic (Using Hands-Free Protocol).
	Student Researcher - IDEAS Lab (Purdue) January 2024 - May 2024 I performed research under Dr. Aniket Bera, developing a classical controller for quadruped walking, specifically to enable the use of motion planning methods that output in cartesian space rather than joint space.
	Student Researcher - CoRAL Lab (Purdue) May 2022 - May 2023 I performed research under Dr. Ahmed H. Qureshi, where I developed methods for multi-agent exploration methods using Reinforcement Learning. This work resulted in a co-first author publication at IROS 2023.

Publications

Efficient Q-Learning over Visit Frequency Maps for Multi-agent Exploration of Unknown Environments

Xuyang Chen*, Ashvin Iyer*, Zixing Wang and Ahmed H. Qureshi
IEEE/RSJ IROS 2023
arXiv / Code / Video
* Indicates Equal Contribution

Projects / Competitions

	Purdue Lunabotics Our team competes in the NASA Lunabotics Competition yearly. In the 2023 season, we placed 6th overall and 3rd in autonomy among all competing teams. During my team on the team, I was the Software Lead, where I led efforts on 1. Utilizing SLAM to map out surrounding objects / obstacles 2. Performing dynamic motion planning (using D*) and robot control (using MPC) 3. Automating subsystems for picking up / depositing objects (using sensor feedback + PID) 3. Handling communication between processors and hardware motor / sensor drivers 4. Creating an entire behavior state machine to handle transitions between different tasks
	Improving Behavior Cloning With Reinforcement Learning I worked with Raghava Uppuluri on utilizing Reinforcement Learning to help improve Beahvior Cloning Policies. Specifically, we looked at improving BC-SAC, a method developed by Waymo. We tried applying their method to a high-dimensional robot arm, and also improved their method by modifying how they integrate BC and RL together.
	Robotics Algorithms I keep a repository of robotics algorithms I implement from scratch, primarily to help improve my understanding (plus they're fun to program :)). I try to update this repository whenever I get spare time.
	Autonomous Robotics Club I joined the Autonomous Robotics Club at Purdue my freshmen year, giving me my first exposure to robotics in college. Here, I tried implementing inverse kinematics and visual servoing for a robotic arm to be able to pick and place chess pieces.