Research

  • Rational Inverse Reasoning

    Ben Zandonati, Tomás Lozano-Pérez, Leslie Pack Kaelbling
    In Submission! 2025
    TL;DR: This paper introduces a framework for few-shot learning from demonstration by inferring the embodied reasoning process behind observed actions. On challenging reasoning tasks, our method achieves strong generalization to new tasks with only 1-3 demonstrations, moving closer towards human performance.
    arXiv

  • Hierarchical Vision-Language Planning for Multi-Step Humanoid Manipulation

    André Schakkal, Ben Zandonati, Zhutian Yang, Navid Azizan
    Robotics Science and Systems (RSS) Workshop on Robot Planning in the Era of Foundation Models 2025
    TL;DR: This work introduces a hierarchical and modular vision-language architecture for multi-step humanoid manipulation tasks. Specifically, we leverage vision-language models (VLMs) for high-level planning and action-chunking transformer (ACT) and whole-body RL for mid and low-level control.
    arXiv Project
  • Image for Investigating Vision Foundational Models for Tactile Representation Learning

    Investigating Vision Foundational Models for Tactile Representation Learning

    Ben Zandonati, Ruohan Wang, Ruihan Gao, Yan Wu
    2023
    TL;DR: This paper explores the use of vision foundational models and pre-trained representations to enhance tactile representation learning and multi-modal continual learning.
    arXiv
  • Image for Towards Optimal Compression: Joint Pruning and Quantization

    Towards Optimal Compression: Joint Pruning and Quantization

    Ben Zandonati, Glenn Bucagu, Adrian Alan Pol, Maurizio Pierini, Olya Sirkin, Tal Kopetz
    2023
    TL;DR: This work presents a method for simultaneously pruning and quantizing neural networks by approximately minimizing the distance in parameter space.
    arXiv
  • Image for FIT: A Metric for Model Sensitivity

    FIT: A Metric for Model Sensitivity

    Ben Zandonati, Adrian Alan Pol, Maurizio Pierini, Olya Sirkin, Tal Kopetz
    International Conference on Learning Representations (ICLR) 2022
    TL;DR: This paper introduces a Fisher Information Metric approximation method for model sensitivity to low-bit quantization.
    arXiv