Dear Editors-in-chief, We are submitting today the paper entitled ‘BeliefNet: A neurosymbolic model to enhance context based traversability predictions for autonomous agents in complex environments’ to the Neurosymbolic AI journal to support your call on neurosymbolic AI in Cyber physical systems. We have conducted this research as part of a PhD looking to use neurosymbolic AI to improve the reasoning capabilities of autonomous systems in complex off route domains. We can confirm that this research is original, has not been published elsewhere and is not being considered elsewhere. The background of the lead author is within Defence, having led in R&D for autonomous systems for the military and was struck by the sheer complexity of complex terrain and how far autonomy was from being achievable. We believe neurosymbolic AI presents an opportunity to transform how autonomous agents could reason and adapt in this complex domain. This premise of this research is to identify a method enabling an agent to determine the risk of a given object based on its context, noting that complex terrain is not linear and individual features do not exist in isolation. In doing so we seek to improve the approximation of human reasoning within autonomous systems. One of the primary constraints of the research is the requirement to enhance trust between the agent and an operator, it is the combination of this constraint and requirement to reason that led us to a neurosymbolic solution. The Belief-Net model, we believe to be a new model structure, based on logical predicates which evolves dynamically to new domains. It is designed as an online learning system, replicating the life-long learning capabilities of humans, it adapts with the agent and enables it to use its prior experience to make judgements on new unseen domains or objects. The model was designed specifically to work in this task, and has some task specific nuances, but it has been expanded to domains external to autonomous systems, showing high performance in other complex reasoning tasks requiring traceability. Outside of its adaptability and learning performance, we see that a key differentiator from some other neurosymbolic models is that the model does not require a-priori logic. Whilst it can be integrated, the model aims to identify its own probabilistic representation of the domain, its beliefs. We have no conflicts of interest to raise and please address any correspondence to tom.scott@cranfield.ac.uk. Thank you for the consideration of our research. Yours sincerely Tom Scott