Research M1 Open IPC
Low-Code Programming and Natural Communication
This module aims to enable process experts to program robots through natural language rather than detailed code or manual block orchestration.
Industrial gap
Existing tools still require knowledge of software libraries, sequence design and optimisation. This can limit who can configure robots and how quickly production changes can be absorbed.
Objective
The proposed workflow connects natural-language task descriptions with formal verification, temporal planning, digital-twin optimisation and online replanning.
How it works
Five-step methodology
- Knowledge-base generation
- PDDL generation and formal verification
- Behavior Trees with STNU temporal uncertainty
- MCTS optimisation in a digital twin
- Execution and online replanning
Key design choices
Human validation
The knowledge base is intended to remain inspectable by domain experts. Human validation helps ensure that generated facts, actions and constraints match the industrial process before they influence execution.
Planning under change
Anytime planning improves candidate solutions while time is available. When disruptive changes occur, rapid heuristic restart provides a faster recovery path instead of waiting for full optimisation.
ActivityA2
Activity A2 covers scientific update, lightweight LLM-based planning, PDDL-to-BT conversion, time-variance modelling, digital-twin optimisation, online replanning, MI.RA/Dexter integration, continuous integration and benchmarking.
Proposed targets
Project target 01
66% programming-time reduction.
Project target 02
Non-pretrained-worker programming demonstration.
Project target 03
Evaluation over multiple planning contexts.
Project target 04
Fleet-management demonstration.
Targets describe planned outcomes, not results already achieved.