The rise of Human-Robot Interaction (HRI) has introduced a new generation of cobots designed to physically assist humans with laborious tasks. To function safely in unpredictable environments, these robots must be autonomous and adaptive. However, modern cobots often use elastic actuators for passive compliance, creating non-linear dynamics that are difficult to manage using traditional control equations.
The Data-Driven Solution
The DLROB project addresses this by shifting from equation-driven to data-driven modeling. Instead of calculating complex dynamics, the system “learns” them through a four-stage Machine Learning (ML) methodology:
- Data Acquisition: Gathering dynamic data from the cobot.
- Neural Network Construction: Building the AI architecture.
- Training & Validation: Refining the model’s accuracy.
- Implementation: Deploying the solution on physical hardware.
Key Objectives & Impact
By leveraging a Model-Based approach—moving from software simulation (MIL/SIL) to hardware implementation (HIL)—DLROB creates precise controllers and simulators for robots with elastic joints. The ultimate goal is to perfect compliance torque control, allowing cobots to transition from industrial factories into daily workplaces and social settings.