Manta-LMPC

Problem

Decentralized agents need to improve repeated motion tasks while preserving separation, obstacle constraints, and a usable route to the goal. Learning from earlier trajectories is only useful when unsafe or incomplete runs are kept out of the learned safe set.

Contribution

The implementation combines artificial-potential-field initialization, decentralized CasADi/IPOPT learning MPC, sampled terminal safe sets, time-to-go costs, and priority-aware spatial hyperplanes.

The broader public testbed also separates heterogeneous ground, surface, and underwater dynamics from guidance, estimation, optimization, and reporting modules.

Outcome

The repository provides a configurable computational research testbed and curated simulation artifacts for repeated multi-agent studies.

This ongoing software project is evaluated through simulation.