Papers

Publications

Selected publications and preprints. Most recent first. See Google Scholar for the full list.

Wiggle and Go! System Identification for Zero-Shot Dynamic Rope Manipulation

Arthur Jakobsson, Abhinav Mahajan, Karthik Pullalarevu, Krishna Suresh, Yunchao Yao, Yuemin Mao, Bardienus Duisterhof, Shahram Najam Syed, Jeffrey Ichnowski

arXiv preprint · 2026

We present a framework that lets robots perform dynamic rope manipulation tasks zero-shot — without large real-world datasets or iterative trial-and-error. A system identification module observes a few seconds of rope motion to predict descriptive physical parameters, which then condition goal-directed action prediction. On a 3D target-striking task we achieve 3.55 cm accuracy compared to 15.34 cm for the no-sysID baseline.

Adversarial Game-Theoretic Algorithm for Dexterous Grasp Synthesis

Yu Chen, Botao He, Yuemin Mao, Arthur Jakobsson, Jeffrey Ke, Yiannis Aloimonos, Guanya Shi, Howie Choset, Jiayuan Mao, Jeffrey Ichnowski

IEEE International Conference on Robotics and Automation (ICRA) · 2026

A game-theoretic formulation of dexterous grasp synthesis where adversarial perturbations during training produce grasps that remain stable under real-world disturbances. Improves contact robustness over standard energy-based grasp planners.

ExpReS-VLA: Specializing Vision-Language-Action Models Through Experience Replay and Retrieval

Shahram Najam Syed^*, Yatharth Ahuja^*, Arthur Jakobsson, Jeffrey Ichnowski
* equal contribution

IEEE International Conference on Robotics and Automation (ICRA) · 2026

ExpReS-VLA specializes general-purpose vision-language-action models to a target task by combining experience replay with retrieval over a small task-specific buffer, narrowing the deployment-gap without full fine-tuning.

arXiv

Work Smarter Not Harder: Simple Imitation Learning with CS-PIBT Outperforms Large-Scale Imitation Learning for MAPF

Arthur Jakobsson^*, Rishi Veerapaneni^*, Kevin Ren, Samuel Kim, Jiaoyang Li, Maxim Likhachev
* equal contribution

IEEE International Conference on Robotics and Automation (ICRA) · 2025

We show that combining a small imitation-learning policy with the CS-PIBT collision-shielding rule outperforms large-scale imitation learning for multi-agent path-finding (MAPF). Strong evidence that careful priors beat raw data scale for combinatorial planning problems.

PDF
Project

Improving Learnt Local MAPF Policies with Heuristic Search

Arthur Jakobsson^*, Rishi Veerapaneni^*, Qian Wang^*, Kevin Ren^*, Jiaoyang Li, Maxim Likhachev
* equal contribution

International Conference on Automated Planning and Scheduling (ICAPS) · 2024

We augment a learnt local multi-agent path-finding policy with classical heuristic search to recover completeness and resolve deadlocks the network alone cannot. The hybrid outperforms either component on standard MAPF benchmarks.