CoPlanVLM, described in a new arXiv preprint, applies a vision language model to a multi-robot planning problem. The system accepts a single natural-language instruction and an image, then generates coordinated paths for two robots. This removes the need for hand-coded task specifications or separate planners for each robot.

The work highlights the potential of VLMs as a flexible interface for robot teams. Because the model interprets both the scene and the command together, it can produce paths that are implicitly coordinated rather than planned independently. The authors position this as a step toward more natural control of multi-robot systems.

As the abstract notes, VLMs offer a flexible, natural-language interface, but the paper does not yet detail performance benchmarks or comparisons with existing methods. The announcement is brief, so further evaluation of reliability and scalability remains open.