On the quest for robotic embodied intelligence - Robots that understand, adapt, and collaborate with us

Aug 25, 2026

Upload by Angel

This document, titled "On the quest for robotic embodied intelligence: Robots that understand, adapt, and collaborate with us" by Georgia Chalvatzaki, explores the dream of intelligent robotic assistants and the challenges in achieving embodied artificial general intelligence (AGI) in the physical world. It delves into how humans learn and adapt through perception-action loops and sensorimotor coordination, questioning if data alone is sufficient for robot learning. The presentation highlights Moravec's Paradox, which states that high-level reasoning is easier to replicate in computers than low-level sensorimotor skills. The document outlines the research focus on bridging assistive robotics with embodied intelligence, moving from understanding human behavior to enabling robots to learn and act effectively among humans. The proposed approach is "Structured Robot Learning," which emphasizes learning with prior knowledge, not just data, incorporating structured representations from geometry and semantics, physics-aware learning, and robust, safe behavior under uncertainty. The work aims to develop modular and composable skills and behaviors for robots, demonstrated through examples like grasping objects anywhere with coarse-to-fine scene understanding, showing multi-level feature rendering for grasp quality and affordance. The timeline indicates progression from PhD research in Assistive Robotics to current work on ROSA2020, Emmy Noether, DFG projects at TU Darmstadt, and the PEARL Lab at TU Darmstadt, focusing on integrating perception and learning for mobile manipulation and structured embodied intelligence.

LanguageEnglish

CategoryTechnology

FormatPDF

Pages28

Reading Time1 h 11 min

AccessFree to download

On the quest for robotic embodied intelligence

The dream of intelligent robotic assistants...

How humans learn and adapt?

Is Data all we need?

Why is embodied intelligence so hard?

From assistive robotics to embodied intelligence

Our approach: Structured Robot Learning

FAQs

What is Moravec's Paradox?

Moravec's Paradox states that it is comparatively easy to make computers exhibit adult-level performance on intelligence tests, but difficult or impossible to give them the skills of a one-year-old, such as sensorimotor skills.

What is the proposed approach for achieving embodied intelligence?

The proposed approach is "Structured Robot Learning," which involves learning with prior knowledge, not just data. It incorporates structured representations from geometry and semantics, physics-aware learning, and aims for robust and safe behavior under uncertainty.

What are the key research areas mentioned in the document?

The key research areas include understanding humans, enabling robots to learn and act among humans, integrating perception and learning for mobile manipulation, and developing structured embodied intelligence.

Aug 25, 2026

Upload by Angel