What the study found
Large language models (LLMs) were used to help assign multi-robot subtasks by combining room-wise object presence probabilities with task decomposition. In the reported experiments, the proposed method achieved 47/50 successful assignments.
Why the authors say this matters
The authors present this as a way to handle instructions that require searching for multiple objects or interpreting context-dependent commands, including underspecified ones such as “Get ready for a field trip.” They suggest the framework can support task decomposition, assignment, sequential planning, and execution.
What the researchers tested
The researchers inferred room-wise object presence probabilities using Bayesian inference with a spatial concept model. They converted these inference results into prompts for LLMs, and used a novel few-shot prompting strategy to help the models infer required objects from ambiguous commands and decompose instructions into subtasks for multiple robots.
What worked and what didn't
In experiments, the proposed method achieved 47/50 successful assignments, compared with 28/50 for random assignment and 26/50 for commonsense-based assignment. The abstract also reports qualitative evaluations with two actual mobile manipulators, which showed the framework could handle underspecified instructions and complete task decomposition, assignment, sequential planning, and execution.
What to keep in mind
The abstract does not provide detailed limitations, error analysis, or the full experimental setup. The reported evidence includes assignment counts and qualitative evaluations with two mobile manipulators, so the available summary is limited in scope.
Key points
- The proposed method achieved 47/50 successful assignments in the reported experiments.
- It outperformed random assignment (28/50) and commonsense-based assignment (26/50).
- The system used Bayesian inference with a spatial concept model to estimate room-wise object presence probabilities.
- A novel few-shot prompting strategy helped LLMs infer required objects from ambiguous commands and split instructions into subtasks.
- Qualitative evaluations with two mobile manipulators showed the framework could handle underspecified instructions such as “Get ready for a field trip.”
Disclosure
- Research title:
- LLM-based planning improved multi-robot task assignments
- Authors:
- Kento Murata, Shoichi Hasegawa, Tomochika Ishikawa, Yoshinobu Hagiwara, Akira Taniguchi, Lotfi El Hafi, Tadahiro Taniguchi
- Institutions:
- Ritsumeikan University, Soka University, Kyushu Institute of Technology, Kyoto University, Kyoto College of Graduate Studies for Informatics
- Publication date:
- 2026-04-27
- OpenAlex record:
- View
Get the weekly research newsletter
Stay current with peer-reviewed research without reading academic papers — one filtered digest, every Friday.

