Talk
MAPF foundations & lifelong planning
Collision constraints, common objectives and the shift from fixed goals to a continuing stream of tasks.
How do you move a fleet through shared space without getting in each other’s way?
Solve this problem with us. You'll use League of Robot Runners Start-Kit and PlanViz to run experiments and leave with a working understanding of MAPF and a way to investigate the design decisions behind multi-agent coordination.
Schedule
This is a half-day programme with short talks, guided experiments and discussion. We start with the fundamentals, then explore the choices that shape a working multi-agent system.
The session date, start time and room are to be announced. Times shown are elapsed from the start of the lab.
Feb 16
- Elapsed time:
0:00–0:30 - Elapsed time:
0:30–0:55Guided lab
Run your first scenario
Get the tools running, visualise a small example, and identify waiting, congestion and task progress.
- Elapsed time:
0:55–1:25Talk
From graph paths to robot motion
Explore orientation, turn actions and motion feasibility, and how they change the meaning of a good path.
- Elapsed time:
1:25–1:55Guided lab
Compare motion models
Adjust motion-related settings and inspect their effect on paths, execution time and congestion.
- Elapsed time:
1:55–2:25Break
Take a break
30 minutes away from the experiments.
- Elapsed time:
2:25–2:55Talk
Task scheduling & execution uncertainty
Connect assignment and pickup-and-delivery decisions to planning, then examine monitoring, delays and replanning.
- Elapsed time:
2:55–3:50Guided lab
Investigate scheduling and delays
Compare scheduling choices and inspect execution traces when agents fall behind their nominal plans.
- Elapsed time:
3:50–4:00Discussion
What did the experiments reveal?
Bring the observations together, discuss emerging research directions, and share questions.
Organisers
Our presenter team brings together Kevin Zheng, Yue Zhang and Jingjin Yu to guide you through the talks, hands-on experiments and discussions.
Contact the organisers
A contact for tutorial enquiries will be added here.
-
Kevin Zheng
Research Engineer, Monash University
Kevin Zheng is a researcher, engineer and teaching staff at Monash University focusing on automated planning and intelligent decision-making. He co-hosted the Optimisation and Planning Summer School in 2025, and maintains tools that serve the planning community, like Posthoc and the MAPF Tracker. His work has been published and recognized with awards at ICAPS 2024 and 2025.
Institution Monash University -
Yue Zhang
Affiliation and biography to be added.
-
Jingjin Yu
Affiliation and biography to be added.
Materials
Find the resources to set up your laptop, follow the guided experiments and revisit the tutorial afterwards. We’ll add links here as the materials are released.
Before the session
Bring a laptop and charger. Once the setup guide is available, follow it and run the test scenario before arriving. You won’t need to implement a solver.
No prior MAPF experience is required. Familiarity with basic AI concepts and some programming or command-line experience will help.
-
Lab repository
Prepared scenarios and configurations for running the experiments.
Link coming soon -
Setup guide and test scenario
Installation steps, the versions used in the lab and a small example to check your setup.
Link coming soon -
Tutorial slides
The talks on MAPF foundations, robot motion, scheduling and execution uncertainty.
Link coming soon -
Exercise notes
Guided experiments, questions to explore and metrics to compare. Optional code-level extensions may also be included.
Link coming soon
Tools and reference projects
The experiments will use the League of Robot Runners Start-Kit to run multi-agent scenarios and PlanViz to inspect plans and execution traces.
These are the upstream projects. Use our setup guide, once released, for the versions and configuration used in the tutorial.
Using the lab materials
Work through the prepared examples alongside the exercise notes. Change a configuration, inspect the resulting paths or execution traces, and compare waiting, congestion, throughput and runtime. Bring your observations and questions to the discussion after each experiment.