Autonomous Robot Swarms with Emergent Adaptive Memes

This work is a collaboration between Chongqing University's Liu Group (experiment-focus) and Princeton University's Austin Group (theory-focus).

"Jeeps" on an interactive display environment -- a 4m x 4m LED lightboard.

Our autonomous robots -- "Jeeps" Mark I version. On a dynamical and interactive 4m x 4m LED lightboard.

A cartoon representing the Jeep world.

A cartoon representing the Jeep world. Upper-left: The "Jeeps" move on a resource landscape generated by an underlying LED light board, which the "Jeeps" sense. Warp drive is the self-generated movement of the "Jeeps" in response to local resource depletion. Center: because of self-generated resource valleys, the "Jeeps" can be collectively trapped when they can form gaseous, solid, liquid and glassy communities due to self-trapping. Bottom center: "Jeeps" mutate their genes which code for resource consumption rates under stress due to loss of resources. Top center: a "Jeep" that cannot evolve fast enough would die and lose its genes. It becomes trapped in the resource, losing the ability to move.  Top right: transfiguration can occur if two motile "Jeeps" donate new genes to the dead "Jeep".

Mark II version of "Jeeps" with the ability to meme.

An idea for "Jeeps" with the ability to meme (exchange ideas and keep knowledges between generations). Mark II version. Currently in devolopment. People familiar with reinforcement learning and neural network architecture are welcome.


  • Wang, G.*, Phan, T.V.*, Li, S., Wombacher, M., Qu, J., Peng, Y., Chen, G., Goldman, D.I., Levin, S.A., Austin, R.H. and Liu, L., 2021. Emergent field-driven robot swarm states. Physical Review Letters126(10), p.108002.
  • Phan, T.V.*, Wang, G.*, Liu, L. and Austin, R.H., 2021. Bootstrapped motion of an agent on an adaptive resource landscape. Symmetry13(2), p.225.

Media Outlets