Proposes a generative design framework for robotic manipulators

Integrates Physical AI to bridge data-driven models with real-world behavior

Key idea:

• Automatically generate manipulator designs from task requirements

Framework:

• Generative design using surrogate models

• Multi-objective optimization: Kinematic and Dynamic performance

• Adaptive sampling for data-efficient learning

Physical AI realization:

• Selected optimal designs are realization as prototypes

• Users can control the manipulator via a mobile application

Contributions:

• Generative Design + Physical AI + Real-world validation

• Establishes an end-to-end autonomous design-to-deployment pipeline

AI-based modular design optimization strategy for large-scale systems

Key idea:

• Replace one-to-one design with modular design (group-based strategy)

Core questions addressed:

• How many groups are needed?

• How many systems per group?

Framework:

• Multi-objective optimization with adaptive grouping strategy

• Identification of cost saturation point

• Iterative search for optimal number of groups

Results:

• Maximized scalability

• Reduced design complexity and cost

• Identification of a “utopia region” balancing performance and efficiency