Skip to content

README.md

Latest commit

 

History

History
127 lines (111 loc) · 6.28 KB

README.md

File metadata and controls

127 lines (111 loc) · 6.28 KB

robotPlanning

This set of simulations illustrates different use cases of the "Planning" idiom. The simulations utilize the HouseExpo](https://github.com/TeaganLi/HouseExpo) dataset and a modified version of their simulator, to generate realistic SLAM output in the form of a map and estimated position.

Before running any of the simulations remember to unpack files in the HouseExpo simulator as stated in the module's README, e.g. by running the command:

cd probMind/examples/misc/HouseExpo/HouseExpo
tar -xvzf json.tar.gz

Exploration

This simulation exemplifies pure exploration by utilizing the "Planning" idiom. The goal of the robot is to explore its environment, i.e., determine the state of all cells in a grid map. The config for this simulation can be found in

probmind/examples/robotPlanning/configs/damgaard2022AKS

and the simulation can be started via the command:

$ cd probmind/examples/robotPlanning
$ conda activate probMind_env_cross_platform
$ python __main__.py -config_file "configs/damgaard2022AKS/configs.yaml" -cpu_cores 1

For more information we refer you to our publication:

@unpublished{damgaard2022AKS,
      title={A Probabilistic Programming Idiom for Active Knowledge Search}, 
      author={Malte R. Damgaard and Rasmus Pedersen and Thomas Bak},
      year={2022},
      note={Submitted}
}

If you like this work or use it in your scientific projects, please use the reference above.

Goal Search

This simulation exemplifies how the "Planning" idiom can be used to find goals in an environment. The simulation builds upon the pure "Exploration" example, by implementing a fallback to exploration whenever the goal is not in sight as illustrated by the GIF above. The config for this simulation can be found in

probmind/examples/robotPlanning/configs/damgaard22GoalSearch

and the simulation can be started via the command:

$ cd probmind/examples/robotPlanning
$ conda activate probMind_env_cross_platform
$ python __main__.py -config_file "configs/damgaard22GoalSearch/configs.yaml" -cpu_cores 1

No-Change Impasse

If the robot only takes action to get closer to its goal as soon it is visible, the robot can end in a "No-Change Impasse". That is a situation in which no progress is made as illustrated in the GIF above. This simulation illustrates how the "Planning" idiom can be used to detect such a "No-Change Impasse" and take actions to escape from it as illustrated in the GIF below.

The config for both of the above simulations can be found in

probmind/examples/robotPlanning/configs/damgaard22Impasses

and the simulation WITH impasse detection can be started via the command:

$ cd probmind/examples/robotPlanning
$ conda activate probMind_env_cross_platform
$ python __main__.py -config_file "configs/damgaard22Impasses/configs.yaml" -cpu_cores 1

For more information we refer you to our publication:

@unpublished{damgaard2022RGS,
      title={Escaping Local Minima Via Appraisal Driven Responses},  
      author={Malte R. Damgaard and Rasmus Pedersen and Thomas Bak},
      year={2022},
      note={Being documented. The Title is tentitative}
}

If you like this work or use it in your scientific projects, please use the reference above.

Planning with Multimodal Action Posterior

In environments with highly non-convex representations, it is often the case that there exists more than one good plan for the future. This simulation illustrates the use of a multimodal action posterior to represent such ambiguities of the best plan for the future. The simulation builds upon the "Goal Search" example. The config for this simulation can be found in

probmind/examples/robotPlanning/configs/damgaard22MultiModalActionPosterior

and the simulation can be started via the command:

$ cd probmind/examples/robotPlanning
$ conda activate probMind_env_cross_platform
$ python __main__.py -config_file "configs/damgaard22MultiModalActionPosterior/configs.yaml" -cpu_cores 1

For more information we refer you to our publication:

@unpublished{damgaard2022MAP,
      title={A Probabilistic Programming Idiom for Planning with Multimodal Action Posterior},  
      author={Malte R. Damgaard and Rasmus Pedersen and Thomas Bak},
      year={2022},
      note={Being documented. The Title is tentitative}
}

If you like this work or use it in your scientific projects, please use the reference above.

Cluster Computing

The "main.py" script has three additional args that can be used to divide the simulation task onto a cluster of computers with different hardware capabilities. The "-cpu_cores" arg determines the number of independent simulations being started on a single computer. The "-thread_start_ID" arg offsets the ID of the simulations being run at a specific computer. The "-total_cluster_threads" arg is used to divide the computational burden evenly. These args implement very basic logic and assume that the processors of each computer are roughly equal fast. As an example, if you want to simulate on a cluster with 3 individual computers with 1 of them having 8 CPU cores and the others having 4 CPU cores, then you can divide the computational burden of a simulation by running the following commands on each pc:

$ python __main__.py -cpu_cores 8 -thread_start_ID 0 -total_cluster_threads 16  # on the pc with 8 cores
$ python __main__.py -cpu_cores 4 -thread_start_ID 8 -total_cluster_threads 16  # on a pc with 4 cores
$ python __main__.py -cpu_cores 4 -thread_start_ID 12 -total_cluster_threads 16  # on a pc with 4 cores

TODO

The following is a list of things that could potentially be improved:

  • Fix the random error happening sometimes in "robot_exploration_Probmind.py" line 70-71
  • implement rotation and FOV in the lidar model
  • Change state to a dict - e.g. z_s_tau["position"]
  • incorporate differentiation between the known location of the goal state and perceived location of the goal state