Like Meshtastic but using Bluetooth so can be done on smartphones (obviously lower range).
Potentially use Gaussian Belief Propagation to generate a routing table. Could reduce size of routing table by only caring about contacts perhaps? Sort of an adaptation of Distributed Spatial Awareness for Robot Swarms
We want super-long range with many, many hops. This means a source is unlikely to know of a complete route to the destination. Instead it must make use of knowledge of the physical world to make educated routing guesses. E.g. each node knows its own location history, and therefore its likeliness to be of help in transmitting a packet. A node would then talk to the previous hop giving a rating of helpfulness, the previous hop may then pick N helpful next hops from its neighbours. N may increase depending on the helpfulness of each option, e.g. it may pick only one next hop if that next hop has high helpfulness, or many next hops if they are all low helpfulness. Each node may hold onto a packet for some TTL duration, during which it may re-transmit if it discovers a more helpful node.
Helpfulness score should also account for frequency of contact with said node.