Human-in-the-Loop Assisted Trafficability Prediction for Planetary Rover on Soft Dangerous Terrains

It needs to predict the soft terrain trafficability of the Mars rover to avoid hazards in emergencies. The current prediction models of Mars terrain trafficability simply transfer information between humans and machines, with lower levels of human-machine interaction fusion, which hinders the creation of effective emergency handling measures. In this paper, a human-in-the-loop multi-indices fusion decision framework called HITL-FD is proposed for Mars rover trafficability prediction. The framework deals with the fuzziness of the trafficability boundaries by fuzzy theory, achieving the fusion processing of multiple evaluation indices and adopting a fuzzy decision tree algorithm to predict trafficability. Additionally, human-assisted decision is incorporated into the decision loop, and a reward function based on decision error is proposed to generate a closed-loop decision framework. The experimental results show that the framework has excellent trafficability prediction performance and reduces 5.1% of MAYBE terrain uncertainty compared to machine-based decision. Note to Practitioners—The trafficability prediction is crucial for Mars rovers traversing soft terrain to avoid emergencies including sinkage. Conventional trafficability prediction models have a low degree of human-machine fusion, which is not beneficial for applications of planetary rovers in practical engineering. In this paper, a multi-indices human-machine fusion decision framework is proposed, specifically designed for predicting the trafficability of Mars rover. In this framework, fuzzy theory is used to address the ambiguity associated with trafficability boundaries. It provides an overall evaluation of trafficability by fusing a single index and the use of a fuzzy decision tree algorithm to make predictions regarding trafficability. Furthermore, the framework incorporates human-assisted decision-making within the closed decision loop, and proposes a reward function based on decision error. Experimental results show that the framework can increase the correctness of trafficability prediction.