On-Demand Robotic Assembly via Differentiable Geometric Part Repair

Key points

• Transitioning from a digital design to a robotic assembly process currently requires months of expert manual tuning to reconcile part geometries with robotic constraints.1
• A generative AI agent translates user prompts into initial 3D geometries, balancing the visual fidelity of the design with select physical constraints.1
• The assemblability of the design is further improved by a gradient-based repair stage that backpropagates through a graph attention network surrogate to adjust component geometries.1
• In addition to correcting for disjointed and overlapping components, we demonstrate hardware-specific corrections, differentiably optimizing the geometry of components to enable robot screwdriving for 86.7% of 60 novel natural language inputs, significantly outperforming prior work by a factor of ten.1

Sources (1)

• [1]On-Demand Robotic Assembly via Differentiable Geometric Part Repair
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 7, 09:57 AM

  This paper presents an end-to-end, autonomous pipeline for the design and physical construction of bespoke wooden assemblies.

  Transitioning from a digital design to a robotic assembly process currently requires months of expert manual tuning to reconcile part geometries with robotic constraints.

Extractive summary: sentences quoted from the sources.