The control system of an autonomous or semi-autonomous machine consists of intelligent perception, world model-ling and decision-making. In this WP, at the application level, we focus on earth moving (wheel loader and excavator), cranes (multi-purpose), and forestry (harvester), and at the technology level, we cover perception and world modelling (for the purpose of map-based localisation, navigation/obstacle avoidance and manipulation/obstacle avoidance), and machine learning (for bucket filling, crane control, and obstacle avoidance).
We will devise robotics control and perception strategies for safe operation of future HDM machinery and for cost effective development and operation in the presence of variability in machines and tasks. Automatic and efficient bucket filling is crucial for full or semi-autonomous (operator assisted) production. Optimised bucket filling does not only affect productivity, but also affects energy consumption, since it is the most fuel-consuming stage of operation. Safety (obstacle detection/avoidance) and autonomous operation depend on robust and versatile perceptions and world modelling in harsh and dynamic environmental conditions, such as construction worksites, outdoor warehouses, and forests. Our objectives are to devise automatic control of underactuated multi-purpose cranes of varying shapes and sizes; devise efficient bucket-filling strategies in the presence of unknown and varying materials to reach comparable productivity as with human experts; world modelling in dynamic outdoor environments for long-term autonomy; and reliable obstacle detection in environments with poor visibility conditions.