README

This repo contains the codes for paper Integrated Task Assignment and Path Planning for Capacitated Multi-Agent Pickup and Delivery.

Compiling

Clone this repo.

$ mkdir build
$ cd build
$ cmake ../
$ make

Dependencies

Boost, Google-sparsehash are required for compiling.

If using ubuntu, you can install them by:

$ sudo apt-get update
$ sudo apt-get install sparsehash
$ sudo apt-get install libboost-all-dev

They are also available on Homebrew, if using mac os.

Usage Examples and Arguments

Kiva Instances are provided in examples folder.

kiva-agent-maps: Including a kiva map with different amount of agents on the map. In each map, "@" indicate obstcale, "." indicate an open space, "e" indicate an endpoint, and "r" indicate an agent initial location. For example, kiva-10-500-5 indicate a map with 10 agents on the map. All maps have same layout, just agents are different.

kiva-tasks: Including 25 kiva task instance with different release frequency. The number in first column is task release timestep, second column is the task starting endpoint ID, the third column is the task goal endpoint ID. For example, the folder with name 0.2-500 indicate each instance have 500 tasks with release frequency of 0.2. For folder without frequency, all tasks are released at timestep 0.

$ ./mapd -m path/to/kiva-agent-maps/kiva-10-500-5.map 
    -a path/to/kiva-agent-maps/kiva-10-500-5.map
    -t path/to/kiva-tasks/1-500/0.task
    -c 60
    -s PP
    --capacity 1
    --objective total-travel-delay
    --only-update-top
    --kiva
    -o path/to/output/file

-m [path] indicate a map to load

-a [path] indicate the agents to load

-t [path] indicate the task to load

-s pp must have this to run coupled task and path planning

--capacity the maximun capacity of agents

--kiva must have this argument to load kiva instances in examples.

--objective [total-travel-delay] The optimize objective, can be total-travel-delay or makespan

--online run example in lifelong mode. Without this argument, codes run in offline mode.

--only-update-top must have this to only update the top elements in the heaps.

--regret run in RMCA. Without this argument codes run as MCA

--anytime enable anytime improvement after assignment.

--group-size [8] group size for anytime improvement.

--destory-method [random] can be "destory-max","multi-max","random"

-c [60] only limit each anytime optimization time. For offline mode, you can set -c as long as possible. For online mode, we only give several seconds for optimze as optimize repeat many times by the assignment process.

Example arguments to recreate algorithms in the paper

MCA

Offline:

--only-update-top --objective total-travel-delay

offline with any time improvement(change numbers and options as your requirements):

--only-update-top --objective total-travel-delay --anytime -c 60 --group-size 5 --destory-method random

Online:

--only-update-top --objective total-travel-delay --online

online with any time improvement(change numbers and options as your requirements):

--only-update-top --objective total-travel-delay --online --anytime -c 1 --group-size 5 --destory-method random

RMCA

Offline:

--regret --only-update-top --objective total-travel-delay

offline with any time improvement(change numbers and options as your requirements):

--regret --only-update-top --objective total-travel-delay --anytime -c 60 --group-size 5 --destory-method random

Online:

--regret --only-update-top --objective total-travel-delay --online

online with any time improvement(change numbers and options as your requirements):

--regret --only-update-top --objective total-travel-delay --online --anytime -c 1 --group-size 5 --destory-method random

Hints

Offline examples does not works well with release frequency smaller than 1, as the low level search(path planning) will be extreme slow if a late released task is assigned frist to pick up.

Seraphli / MCA-RMCA