Publications

Distributed Upload and Active Labeling for Resource-Constrained Fleet Learning CoRL, 2025
Oguzhan Akcin, Harsh Goel, Ruihan Zhao, Sandeep Chinchali
We propose a decentralized two-stage framework for data collection in multi-robot fleets, achieving up to 31.1% accuracy gains and 13% higher real-world task success rates.

Fleet Supervisor Allocation: A Submodular Maximization Approach CoRL, 2024
Oguzhan Akcin, Ahmet Ege Tanriverdi, Kaan Kale, Sandeep Chinchali
We created an allocation algorithm to collect data from robots through human supervision.

Time Weaver: A Conditional Time Series Generation Model (Spotlight) ICML, 2024
Sai Shankar Narasimhan, Shubhankar Agarwal, Oguzhan Akcin, Sujay Sanghavi, Sandeep Chinchali
We created a new conditional diffusion model that is tailored for time-series data generation.

ControlPay: An Adaptive Payment Controller for Blockchain Economies ICB, 2024
Oguzhan Akcin, Robert P. Streit, Benjamin Oommen, Sriram Vishwanath, Sandeep Chinchali
We model infrastructure token economies as a dynamical system and utilize modern control methods to optimize the overall decentralized network’s performance.

Fleet Active Learning: A Submodular Maximization Approach CoRL, 2023
Oguzhan Akcin, Orhan Unuvar, Onat Ure, Sandeep Chinchali
We created a general framework for applying active learning methods to distributed robot fleets and showed its ability to achieve the same optimality bound as a centralized approach.

Decentralized Data Collection for Robotic Fleet Learning: A Game-Theoretic Approach CoRL, 2022
Oguzhan Akcin, Po-han Li, Shubhankar Agarwal, Sandeep Chinchali
We model the data collection problem from a fleet of robots as a potential game between N-robots and a decentralized method that achieves an optimal solution.

Decentralized Sharing and Valuation of Fleet Robotic Data CoRL, 2021
Yuchong Geng, Dongyue Zhang, Po-han Li, Oguzhan Akcin, Ao Tang, Sandeep Chinchali
We propose a decentralized learning framework with a decentralized voting and feedback mechanism to prioritize the data that is of high value to other robots.