← Back ICRA 2026

DYNAMO: Dependency-Aware Deep Learning Framework for Articulated Assembly Motion Prediction

Mayank Patel, Rahul Jain, Asim Unmesh, Karthik Ramani

PDF

AI summary

Key figure (auto-extracted from paper)

DYNAMO accurately predicts interdependent rigid-body motion in complex gear assemblies directly from static CAD geometry, outperforming existing baselines.

Coupled motion prediction Gear assemblies Dependency-aware learning CAD motion inference Graph neural networks MechBench dataset

Problem

Existing methods struggle to predict motion in mechanical assemblies because they rely on predefined joints or assume independent part mobility, failing to capture how motion propagates through geometric coupling and contact.

Approach

The authors introduce MechBench, a dataset of 693 synthetic gear assemblies, and DYNAMO, a neural network that combines a coupling-aware graph neural network with a temporal decoder to predict per-part 3D motion trajectories directly from segmented CAD point clouds without joint annotations.

Key results

MechBench dataset of 693 diverse synthetic gear assemblies with ground-truth motion trajectories
DYNAMO architecture integrating PointNet++, coupling-aware GNN, and temporal decoder
Superior accuracy and temporal consistency over RigidNet-LSTM, RigidNet-Transformer, and RPM-Net baselines
Novel dependency-aware framework for data-driven learning of coupled mechanical motion in CAD

Why it matters

Enables robots and design automation tools to infer physically valid, interdependent part motions directly from static CAD models, advancing assembly automation and robotic manipulation.

Abstract

Understanding the motion of articulated mechani- cal assemblies from static geometry remains a core challenge in 3D perception and design automation. Prior work on everyday articulated objects such as doors and laptops typically assumes simplified kinematic structures or relies on joint annotations. However, in mechanical assemblies like gears, motion arises from geometric coupling, through meshing teeth or aligned axes, making it difficult for existing methods to reason about relational motion from geometry alone. To address this gap, we introduce MechBench, a benchmark dataset of 693 di- verse synthetic gear assemblies with part-wise ground-truth motion trajectories. MechBench provides a structured setting to study coupled motion, where part dynamics are induced by contact and transmission rather than predefined joints. Building on this, we propose DYNAMO, a dependency-aware neural model that predicts per-part SE(3) motion trajectories directly from segmented CAD point clouds. Experiments show that DYNAMO outperforms strong baselines, achieving accu- rate and temporally consistent predictions across varied gear configurations. Together, MechBench and DYNAMO establish a novel systematic framework for data-driven learning of coupled mechanical motion in CAD assemblies. Project page: https://dynamo-web.pages.dev/

Index terms

Deep Learning for Visual Perception Data Sets for Robotic Vision Visual Learning