Kaushik Shivakumar

I am a 5th year MS student at UC Berkeley's AUTOLab, directed by Professor Ken Goldberg, where I work on robotics research related to deformable object manipulation, leveraging language models, and reinforcement learning.

Email  /  Résumé  /  Google Scholar  /  Twitter  /  Github  /  LinkedIn

I'm interested in developing broadly intelligent robotic systems that can assist humans, automate difficult and tedious tasks, and aid humans in pushing society forward. Below are some of the research papers I have had the chance to work on and publish so far.

Leveraging uncertainty quantification to actuate interactive perception primitives to autonomously untangle long cables.

Conducted a study of cloud-based robotics research, where we developed novel algorithms based on behavior cloning to smooth and fold real garments with one arm.

A study on how to best combine ML-based perception and motion primitives to perform a very challenging deformable manipulation task of untangling long cables.

In this work, we study how we can use pretrained ensembles – encouraged via a KL-divergence in their loss function to be as diverse as possible – to then generalize to new tasks using Reinforcement learning.

Below are some other/older research projects I've worked on.

Studied the problem of autonomously identifying and navigating terrain using visual features to map to driveability characteristics.

Using deep learning for organ segmentation from 3D CT-scans of patients.

Used large-scale county datasets to identify the demographic and environmental factors that correlate and may cause cancer incidences across the United States.