Pedestrian Simulator (Pedsim) is a 2D pedestrian simulator based on the social force model of Helbing et. al. This repository is an extension of the original Pedsim repository and also incorporating features from Ronja Güldenring's fork. It's supposed to be used in conjunction with Arena-Rosnav.
Movement of the agents in the simulation is controlled by these forces:
- desired force (pulls agent towards the current destination)
- social force (pushes agent away from other agents)
- obstacle force (pushes agent away from obstacles)
- robot force (pushes agent away from robot)
Additionally agents can perform different actions that interrupt or modify the movement like talking to each other, stand as a group or take a walk with another agent. We also implemented the possibility to spawn a forklift agent that has the same basic movement functionality but has different actions e.g. loading/unloading objects.
For Pedsim to work together with Arena-Rosnav it needs to be synced to Flatland Simulator. This is done via a flatland plugin that subscribes to the simulated_agents topic published by Pedsim and updates the position of the corresponding flatland models accordingly.
 |
 |
|---|---|
| Human agents walking together | Human agents standing as a group |
 |
|---|
| Forklift agent doing some work |
cd ~/catkin_ws/src/arena-rosnav
git checkout sim_to_real
rosws update --delete-changed-uris .
cd ~/catkin_ws
catkin_make -DPYTHON_EXECUTABLE=/usr/bin/python3
source devel/setup.zsh
workon rosnav
roslaunch arena_bringup pedsim_demo.launch
This will start Flatland, Pedsim and a script that will spawn a few agents.
Copy this to your .rosinstall file:
- git:
local-name: ../forks/pedsim_ros
uri: https://github.com/ignc-research/pedsim_ros
version: master
Afterwards run rosws update
cd ~/catkin_ws/src/arena-rosnav
rosws update .
Add the following to the launch file you want to use:
<arg name="simulation_factor" default="1"/>
<arg name="update_rate" default="25.0"/>
<arg name="scene_file" default="$(find simulator_setup)/pedsim_scenarios/empty.xml"/>
<node name="pedsim_simulator" pkg="pedsim_simulator" type="pedsim_simulator" output="screen">
<param name="simulation_factor" value="$(arg simulation_factor)" type="double"/>
<param name="update_rate" value="$(arg update_rate)" type="double"/>
<param name="scene_file" value="$(arg scene_file)" type="string"/>
</node>
Pedsim was originally meant to be started using a scenario.xml file. In our version you can spawn agents dynamically but you still need to provide a scenario.xml file. Best practice is to just use an empty scenario. For the above snippet to work you need a file called empty.xml in this folder:
~/catkin_ws/src/arena-rosnav/simulator_setup/pedsim_scenarios/
empty.xml should contains this:
<?xml version="1.0" encoding="UTF-8"?>
<scenario>
</scenario>
After launching, the pedsim_simulator node should be running and advertising its services.
Spawns a number of agents in Pedsim. Additionally it spawns models in Flatland by calling the /spawn_models service. Uses an array of Ped messages. Fields that can be filled:
| Type | Name | Description | typical value |
|---|---|---|---|
| int16 | id | agent id(*) | - |
| geometry_msgs/Point | pos | starting position | - |
| string | type | Agent Type | "adult" |
| int16 | number_of_peds | number of agents in this cluster | 1 |
| float64 | vmax | max. velocity | 1.4 |
| float64 | chatting_probability | Probability entering StateTalking | 0.01 |
| float64 | tell_story_probability | Probability entering StateTellStory | 0.01 |
| float64 | group_talking_probability | Probability entering StateGroupTalking | 0.01 |
| float64 | talking_and_walking_probability | Probability entering StateTalkingAndWalking | 0.01 |
| float64 | requesting_service_probability | Probability entering StateRequestingService | 0.01 |
| float64 | max_talking_distance | how close agents need to be, to be able to talk to them | 1.5 |
| float64 | max_servicing_radius | radius in which the service robot will provide service | 10.0 |
| float64 | talking_base_time | time agent spends being in StateTalking (***) | 10.0 |
| float64 | tell_story_base_time | time agent spends being in StateTellStory (***) | 10.0 |
| float64 | group_talking_base_time | time agent spends being in StateGroupTalking (***) | 10.0 |
| float64 | talking_and_walking_base_time | time agent spends being in StateTalkingAndWalking (***) | 10.0 |
| float64 | receiving_service_base_time | time agent spends being in StateReceivingService (***) | 30.0 |
| float64 | requesting_service_base_time | time agent spends being in StateRequestingService (***) | 30.0 |
| float64 | force_factor_desired | desired force will be multiplied by this factor | 1.0 |
| float64 | force_factor_obstacle | obstacle force will be multiplied by this factor | 2.0 |
| float64 | force_factor_social | social force will be multiplied by this factor | 2.0 |
| geometry_msgs/Point[] | waypoints | destinations for the agent | - |
| int16 | waypoint_mode | can be loop (0) or random (1) | 1 |
| string | yaml_file | path to flatland model file (**) | - |
(*) not used currently; agents automatically increment their id, starting at 1
(**) models are found in arena-rosnav/simulator_setup/dynamic_obstacles/
(***) this value will be multiplied by a random factor (0.5 to 1.5) every time the state gets activated
Resets all agents to their starting position and resets their current destination. Uses std_srvs/Trigger Message.
Removes all agents from Pedsim and also removes their corresponding models from Flatland. Uses std_srvs/SetBool Message.
Calls remove_all_peds first and then spawn_peds with the given array of Ped messages.
Spawns a number of interactive obstacles. Agents will interact with these when they get close. Additionally it spawns models in Flatland by calling the /spawn_models service. Uses an array of InteractiveObstacle messages. Fields that can be filled:
| Type | Name | Description | typical value |
|---|---|---|---|
| string | type | obstacle type, only "shelf" available currently | "shelf" |
| geometry_msgs/Pose | pose | position and angle | - |
| float64 | interaction_radius | if the appropriate agent enters this radius, it will interact | 4.0 |
| string | yaml_file | path to flatland model file (*) | - |
(*) models are found in arena-rosnav/simulator_setup/dynamic_obstacles/
Removes all interactive obstacles from Pedsim and also removes their corresponding models from Flatland. Uses std_srvs/Trigger Message.
Calls remove_all_interactive_obstacles first and then spawn_interactive_obstacles with the given array of InteractiveObstacle messages.
The most important topic is /pedsim_simulator/simulated_agents. It publishes an array of AgentState messages. The most important fields are:
| Type | Name | Description |
|---|---|---|
| string | type | Agent Type |
| string | social_state | Agent State |
| geometry_msgs/Pose | pose | position |
| geometry_msgs/Twist | twist | velocity |
| float64 | direction | angle the agent is facing |
Note: pose contains information about orientation, just like direction does. direction is meant to be the angle, the agent is facing on a semantic level. Due to how Pedsim works, the orientation in pose can change very rapidly based on small movements.
Agents currently can be one of four types which affects their behavior
| ID | Name |
|---|---|
| 0 | adult |
| 1 | child |
| 2 | elder |
| 3 | vehicle |
| 4 | servicerobot |
The agents behavior is controlled by a statemachine with the following states depending on agent type:
Standing still.
Walking to the next destination.
Like StateWalking but increased movement speed.
Talking to another agent, standing still.
When entering this state the agent will stand still. Listeners will gather around in a circular pattern.
When entering this state the agent triggers a group talk with the agent itself and other agents in range gathering in a cirular pattern.
This state will be triggered whenever one of these is true:
- another agent is talking to this agent directly (StateTalking)
- another agent in close proximity is telling a story (StateTellStory)
- another agent in close proximity has initiated a group talk (StateGroupTalking)
Talk to another agent while walking to the next destination. This means that the listening agent will be walking directly next to this agent.
This state will be triggered when another agent entered StateTalkingAndWalking with this agent as the recipient. This agent will walk directly next to the talking agent.
When entering this state the agent will slow down and signal to service robots that it's requesting service. If no service robot reached this agent in a given time it switches back to StateWalking.
This state will be activated once a service robot has reached this agent. Switch back to StateWalking after the configured time has passed.
Driving to the next destination.
Reached the destination (shelf) and turn towards it.
Standing still.
Standing still.
Standing still.
Driving to the next destination.
Activated once there is an agent in state StateRequestingService in a circular area around the service robot. Radius of the area can be configured by the parameter max_servicing_radius in pedsim_msgs/Ped. The Service robot will now drive towards the requesting agent.
Activated once the service robot has reached the requesting agent. Will stay in this state as long as the other agent is in StateReceivingService.