serialization the data and save them

[feixiao5566]

hello Mryellow, when i fork your code I notice that your vision keeps the serialization, so I guess you want to save the map data, but you don`t finish that. Here is my program, hope that could be useful:^)
main_lv.cpp

#include <fstream>

  ros::spin();

  //add this four line between  ros::spin(); and return 
  std::ofstream ofs( "store.dat" );
  boost::archive::text_oarchive ar(ofs);
  ar & lv;
  ofs.close();

  return 0;

your code helps me a lot, thankyou!

[mryellow]

Not mine originally and I'm not completely across the code so not sure what the intentions are in places.

With a bit more work it could be rather functional. I like the general organisation of the code, easy enough to read.

[mryellow]

What I'm doing with that goal order reversal is experimenting with training DQN to make good Ratslam map topology. Became a bit busy on other stuff but had some promising results with the DQN side. Haven't gone to actor-critic, actor-learner style models or any LTSM implementation which would enable bypassing walls to approach goals, but with a goal placed not far in the past walls shouldn't be too much of an issue.

The idea is to place a goal behind the agent and by turning around have Ratslam develop experiences in both directions (without omni-direction offset feature), thinking rats themselves do this running back and forth a lot. Hoped it would strengthen those paths and connect nodes rather than gradually becoming a mess.

https://github.com/mryellow/ros-dqn-rat-js
https://github.com/mryellow/reinforcejs/tree/demo-multiagent