PoultryBot, a robot for poultry house applications: localisation, path planning, object recognition and performance evaluation

Nowadays, the aviary system is the most common housing system for laying hens in the Netherlands. The large behavioural freedom for the hens in this housing system contributes positively to hen welfare. On the other hand, this led to an increase in the need for human labour, for example to manually collect floor eggs and monitor the complex interactions in the flock. Given the increased attention for using robots, also in agriculture, to replacing undesired labour and increase sensing opportunities, using an autonomous robot for floor egg collection is proposed.

As such robot did not exist prior to this research, several required components were evaluated in more detail on their performance for poultry house applications. First, probabilistic localisation was evaluated in an empty poultry house, to determine its applicability for such repetitive environments and to find the most suitable algorithm settings. Next, path planning for floor egg collection requires coverage of the whole house, while revisiting behaviour during the day should depend on probability of floor egg occurrence. For this, the non-uniform repetitive area coverage (NURAC) path planner was developed. To discriminate between objects in the poultry house, such as hens, housing and eggs, a simple method for image classification based on spectral reflectance properties was presented. Integrating these components with a navigation method and a floor egg collection device resulted in PoultryBot, the world’s first autonomous mobile robot for poultry houses. Experimental evaluation showed PoultryBot to be capable of autonomous navigation and floor eggs collection under practical conditions with live hens. Finally, Technology Readiness Levels were defined for the subsystems and PoultryBot as whole based on the achieved result. From that, indications are given on what is required to advance on the TRL scale, and which approaches seem useful given the experiences so far.