Omkar Patil

Hi there! I’m a PhD student @ ASU working in the Logos Robotics Lab under the guidance of Dr. Nakul Gopalan. My main focus area is compositional robot learning, where I try to develop methods that utilize the principle of compositionality to learn skills efficiently. Prior to joining ASU, I worked for 3 years at Wells Fargo where I did a variety of research in NLP. In my previous life, I did my undergraduate and master’s from IIT Madras (India), focusing on Robotics. My master’s thesis was done under the guidance of Dr. Anurag Mittal.

Apart from my work, I love hiking and have done several extensive hikes in the Himalayas and plan to do some more in the US! I have a strong liking for landscapes and nature.

news

Oct 21, 2024	My paper Few-shot Compositional Learning of Motions has been accepted for Compositional Learning Workshop @ NeurIPS 2024!
Oct 01, 2024	Two co-authored papers Learning Temporally Composable Task Segmentations with Language and Hardware-Software Co-Design for Path Planning by Drones have been accepted for IROS 2024! I will also presenting these works in Abu Dhabi.
Aug 01, 2023	Joined the PhD program at ASU!

latest posts

May 10, 2024	Explaining Life from the Lens of Reinforcement Learning
Mar 31, 2020	COVID Contact Tracing Explained — Corona Kavach

selected publications

Composing Diffusion Policies for Few-shot Learning of Movement Trajectories

Omkar Patil, Anant Sah, and Nakul Gopalan

2024

Abs Bib

Humans can perform various combinations of physical skills without having to relearn skills from scratch every single time. For example, we can swing a bat when walking without having to re-learn such a policy from scratch by composing the individual skills of walking and bat swinging. Enabling robots to combine or compose skills is essential so they can learn novel skills and tasks faster with fewer real world samples. To this end, we propose a novel compositional approach called DSE- Diffusion Score Equilibrium that enables few-shot learning for novel skills by utilizing a combination of base policy priors. Our method is based on probabilistically composing diffusion policies to better model the few-shot demonstration data-distribution than any individual policy. Our goal here is to learn robot motions few-shot and not necessarily goal oriented trajectories. Unfortunately we lack a general purpose metric to evaluate the error between a skill or motion and the provided demonstrations. Hence, we propose a probabilistic measure - Maximum Mean Discrepancy on the Forward Kinematics Kernel (MMD-FK), that is task and action space agnostic. By using our few-shot learning approach DSE, we show that we are able to achieve a reduction of over 30% in MMD-FK across skills and number of demonstrations. Moreover, we show the utility of our approach through real world experiments by teaching novel trajectories to a robot in 5 demonstrations.
@misc{patil2024composingdiffusionpoliciesfewshot, title = {Composing Diffusion Policies for Few-shot Learning of Movement Trajectories}, author = {Patil, Omkar and Sah, Anant and Gopalan, Nakul}, year = {2024}, eprint = {2410.17479}, archiveprefix = {arXiv}, primaryclass = {cs.RO}, url = {https://arxiv.org/abs/2410.17479}, }