A joint research project by AIT, Bosch Rexroth AG and Robert Bosch GmbH addresses this challenge with a new approach to controlling hydraulic excavators. The aim is to develop standardised, machine-independent functional software for assistance and automation functions in hydraulic excavators.

The core idea is a uniform interface that ‘translates’ the manufacturer-specific characteristics of the hydraulics, such as the interaction of oil, pressure and valves, into a common format. This creates a kind of common language for motion commands, in which assistance systems and autonomous functions can access the same clearly defined interface. This abstraction is a decisive step towards scalability: autonomous functions and automated motion sequences can be more easily transferred to different machine types and, in the long term, deployed across entire machine fleets.

The technical foundation is a hybrid, physics-informed AI architecture. A physics-based base model describes machine behaviour using minimal data and ensures stability as well as predictable and traceable system behaviour. This is supplemented by an online learning AI model that compensates for deviations in real time and continuously adapts to changing operating conditions – such as varying loads, ground conditions or component wear.

The approach was tested on a 12-tonne hydraulic excavator under real-world operating conditions. In typical working scenarios such as levelling flat surfaces and slopes (e.g. road verges), the deviation from the desired movement was significantly reduced by the online-trained AI model: by up to 60 % on flat surfaces and up to 70 % on slopes. As the control deviations are in the range of a few centimetres, the requirements of the construction industry were thus met.

The results show that a standardised interface for hydraulic excavators is technically feasible and represents a key enabler for automation. By combining a physical model with an AI model adapted online, robust, adaptive and transferable control concepts can be realised.

In the long term, this approach opens up new prospects for more productive, safer and more automated construction sites – from the use of intelligent assistance systems to fleets of autonomously operating machines.