Agent Function Conditions (Conditional Behaviours)

Agent function conditions are primarily used to split agent populations, allowing them to diverge between agent states. As such, when creating an AgentFunctionDescription the initial and end states for the executing agent should be specified if using more than the default agent state. Following this, all agents in the specified initial state will execute the agent function and move to the end state. In order to allow agents in the same state to diverge an agent function condition must be added to the function.

Agent function conditions are executed by all agents in the input state before the main agent function, and each agent must return either true or false. Agents which return true pass the condition and continue to execute the agent function and transition to the end state, agents which return false fail the condition, do not execute the agent function and remain in the input state.

Agent function conditions are specified using the FLAMEGPU_AGENT_FUNCTION_CONDITION in C++, this differs from the normal agent function macro as only the function condition name must be specified. Within Python, agent function conditions are specified using the @agent_function_condition attribute and then translated to C++ using the code generator. The Python function must have a return type annotation of bool and is not permitted to have any function arguments.

Within agent function conditions a reduced read-only Device API, ReadOnlyDeviceAPI, is available. This only permits reading agent variables, reading environment variables and random number generation.

Example definition of an agent function condition:

// This agent function condition only allows agents who's 'x' variable equals '1' to progress
    return FLAMEGPU->getVariable<int>("x") == 1;

Much like regular agent functions the Python API must use the RTC interface so that the agent function conditions can be compiled at runtime.

Example definition of how the above agent function condition would be attached to an agent function:

// A model is defined
ModelDescription m("model");
// It contains an agent with 'variable 'x' and two states 'foo' and 'bar'
AgentDescription a = m.newAgent("agent");
// The agent has an agent function which transitions agents from state 'foo' to 'bar'
AgentFunctionDescription af1 = a.newFunction("example_function", ExampleFn);
// Only agents that pass function condition 'x_is_1' may execute the function and transition


If you wish to store C++ RTC agent function conditions in separate files setRTCFunctionCondition() can be replaced with setRTCFunctionConditionFile(), instead passing the path to the agent function conditions source file (relative to the working directory at runtime). This will allow them to be developed in a text editor with C++/CUDA syntax highlighting.