此程序基于以下文章：

• "SIPP: Safe Interval Path Planning for Dynamic Environments",
• "Complete decentralized method for on-line multi-robot trajectory planning in well-formed infrastructures",
• "Lifelong Multi-Agent Path Finding for Online Pickup and Delivery Tasks",
• "Persistent and Robust Execution of MAPF Schedules in Warehouses".

此程序的运行效果可参考视频：

• v.youku.com/v_show/id_XNDQ2MTgzNzczMg==.html

1. Compile. In the base directory of this project, do "mkdir build && cd build && cmake .. && make".
2. Make sure redis server is started, and there is no key named "ks". With Ubuntu, to start redis, run "redis-server", then do "echo 'del ks' | redis-cli".
3. Start the python GUI by "python3 ../src/manual_utilities/ks_gui.py". If the module "tk" was not installed, try pip3.
4. Start the main program by "./bin/local_test_main"(the first round of computation may take 15 seconds).

1. In this project, communication between modules is written as function call(receiver maintains a message queue and a lock), as compared to use RPC systems like Stubby. The advantage of this approach is lighter weight, the disadvantage is having this non-standard setup needs more reasoning.

2. For all message queue, the sender can send in an async way, the receiver should copy and process.

3. Computing the action dependency graph is time consuming(for 300 robots, it takes around 10s). Move it out of the critical path makes the program more responsive.

1. Components in a large program should be tested separately. To achieve this, APIs between components need to be defined. There needs to be an API class and an implementation class.

2. Should finish all behaviors before turn on -O2, since with optimization turned on, it is much harder to debug.