State Machines

When designing operant behaviour assays in systems neuroscience, it is useful to describe the task as a sequence of states the system goes through (e.g. stimulus on, stimulus off, reward, inter-trial interval, etc). Progression through these states is driven by events, which can be either internal or external to the system (e.g. button press, timeout, stimulus offset, movement onset). It is common to describe the interplay between states and events in the form of a finite-state machine diagram, or graph, where nodes are states, and arrows are events.

For example, a simple reaction time task where the subject needs to press a button as fast as possible following a stimulus is described in the following diagram:

stateDiagram-v2
    direction LR
    [*] --> ITI
    ITI --> ON: elapsed
    ON --> Reward: hit
    ON --> Fail: miss
    Reward --> [*]
    Fail --> [*]

The task begins with an inter-trial interval (ITI), followed by stimulus presentation (ON). After stimulus onset, advancement to the next state can happen only when the subject presses the button (success) or a timeout elapses (miss). Depending on which event is triggered first, the task advances either to the Reward state, or Fail state. At the end, the task goes back to the beginning of the ITI state for the next trial.

The exercises below will show you how to translate the above diagram of states and events into an equivalent Bonsai workflow, which can be easily adapted and modified to describe many different operant behaviour tasks.

Exercise 1: Declaring and logging external hardware events

In this worksheet, we will be using an Arduino or a camera as an interface to detect external behaviour events. For experimental purposes, it is very helpful to record and timestamp all of these events, independently of which state the task is in.

• Connect a digital sensor (e.g. beam-break, button, TTL) into Arduino pin 8.
• Insert a DigitalInput source and set it to Arduino pin 8.
• Insert a PublishSubject operator and set its Name property to Response.
• Insert a Timestamp operator.
• Insert a CsvWriter sink and configure its FileName property with a file name ending in .csv.
• Run the workflow and activate the digital sensor a couple of times. Stop the workflow and confirm that the events were successfully timestamped and logged in the .csv file.

Note

In order to avoid hardware side-effects, it is highly recommended to declare all hardware connections at the top-level of the workflow, and interface all trial logic using subject variables. This will have the added benefit of allowing for very easy and centralized replacement of the rig hardware: as long as the new inputs and configurations are compatible with the logical subjects, no code inside the task logic will have to be changed at all.

• Right-click the DigitalInput source, select Create Source (bool) > BehaviorSubject, and set its Name property to Led.
• Insert a DigitalOutput sink and set it to Arduino pin 13.

Exercise 2: Inter-trial interval and stimulus presentation

Translating a state machine diagram into a Bonsai workflow begins by identifying the initial state of the task (i.e. the beginning of each trial). It is often convenient to consider the inter-trial interval period as the initial state, followed by stimulus presentation.

• Insert a Timer source and set its DueTime property to be about 3 seconds.
• Insert a Sink operator and set its Name property to StimOn.
• Double-click on the Sink node to open up its internal specification.

Note

The Sink operator allows you to specify arbitrary processing side-effects without affecting the original flow of events. It is often used to trigger and control stimulus presentation in response to events in the task. Inside the nested specification, Source1 represents input events arriving at the sink. In the specific case of Sink operators, the WorkflowOutput node can be safely ignored.

StimOn:

• Insert a Boolean operator following Source1 and set its Value property to True.
• Find and right-click the Led subject in the toolbox and select the option Multicast.
• Run the workflow a couple of times and verify that the sequence of events is progressing correctly.

Note

Opening a new connection to the Arduino can take several seconds due to the way the Firmata protocol is implemented. This may introduce a slight delay in starting the task. This delay is only present at the start of execution and will not affect the behavior of the state machine.

• In the main top-level workflow, insert a Delay operator and set its DueTime property to a couple of seconds.
• Copy the StimOn operator and insert it after the Delay (you can either copy-paste or recreate it from scratch).
• Rename the new operator to StimOff and double-click it to open up its internal representation.
• Set the Value property of the Boolean operator to False.
• Run the workflow a couple of times. Is it behaving as you would expect?
• Insert a Repeat operator after the StimOff.
• Run the worklow. Can you describe in your own words what is happening?
• Optional: Draw a marble diagram for Timer, StimOn, Delay, and Repeat.

Exercise 3: Driving state transitions with external behaviour events

• Delete the Delay operator.
• Insert a SelectMany operator after StimOn, and set its Name property to Response.
• Double-click on the SelectMany node to open up its internal specification.

Note

The SelectMany operator is used here to create a new state for every input event. Source1 represents the input event that created the state, and WorkflowOutput will be used to report the end result from the state (e.g. whether the response was a success or failure).

Response:

• Subscribe to the Response subject in the toolbox.
• Insert a Boolean operator and set its Value property to True.
• Insert a Take operator and set its Count property to 1.
• Delete the Source1 operator.
• Connect the Boolean operator to WorkflowOutput.
• Run the workflow a couple of times and validate the state machine is responding to the button press.

Exercise 4: Timeout and choice

Response:

• Inside the Response node, insert a Timer source and set its DueTime property to be about 1 second.
• Insert a Boolean operator and set its Value property to False.
• Join both Boolean operators with a Merge combinator.
• Connect the output of Take to WorkflowOutput.
• Run the workflow a couple of times, opening the visualizer of the Response node.

Describe in your own words what the above modified workflow is doing.

Exercise 5: Specifying conditional task outcomes

• Insert a Condition operator after the StimOff node, and set its Name property to Success.
• In a new branch from StimOff, insert another Condition, and set its Name property to Miss.
• Double-click on the Condition operator to open up its internal specification.

Note

The Condition operator allows you to specify arbitrary rules for accepting or rejecting inputs. Only inputs which pass the filter specified inside the Condition are allowed to proceed. It is often used to represent choice points in the task. Inside the nested specification, Source1 represents input events to be tested. The WorkflowOutput node always needs to be specified with a bool input, the result of whether the input is accepted (True) or rejected (False). Usually you can use operators such as Equal,NotEqual,GreaterThan, etc for specifying such tests.

Miss:

• Insert a BitwiseNot operator after Source1.

Why did we not need to specify anything for the Success condition?

• In the top-level workflow, insert a SelectMany operator after the Success condition and change its Name property to Reward.
• Inside the Reward node you can specify your own logic to signal the trial was successful. For example, you can make the LED blink three times in rapid succession:

Reward:

• Insert a Timer node and set both the DueTime and the Period properties to 100ms.
• Insert a Mod operator and set the Value property to 2.
• Insert the Equal operator and leave its Value property at 0.
• Find and right-click the Led subject in the toolbox and select the option Multicast.
• Insert a Take operator and set the Count property to 6.
• Insert the Last operator.

Try out your state machine and check whether you understand the behavior of the reward signal.

• Copy the Reward node, paste it after the Miss condition, and change its Name property to Fail.

• Optional: Modify the Fail state in some way to signal a different trial outcome (e.g. make the LED blink more times, or move a motor).

• In the top-level workflow, insert a Merge operator and connect to it the outputs of both conditional branches and before the Repeat node.

Try out your state machine and introduce variations to the task behavior and conditions.

Exercise 6: Go/No-Go task

Implement the following trial structure for a Go/No-Go task.

stateDiagram-v2
    direction LR
    NoGo: No-Go
    FalseAlarm: False<br>Alarm
    CorrectReject: Correct<br>Reject
    [*] --> ITI
    ITI --> Go: 50%
    ITI --> NoGo: 50%
    Go --> Hit
    Go --> Miss
    NoGo --> FalseAlarm
    NoGo --> CorrectReject

• Trials should be sampled from a uniform distribution using the Numerics package (install from Tools > Manage Packages).
• Response events should be based on a button press, and reject events on a timeout.
• Make sure to implement different visual or auditory feedback for either the cue or reward/failure states.

Tip

To sample values from a discrete uniform distribution, you can use the following workflow:

• Record a timestamped chronological log of trial types and rewards into a CSV file using a BehaviorSubject.

Exercise 7: Conditioned place preference

Implement the following trial structure for conditioned place preference. enter and leave events should be triggered in real-time from the camera, by tracking an object moving in or out of a region of interest (ROI). Reward should be triggered once upon entering the ROI, and not repeat again until the object exits the ROI and the ITI has elapsed.

stateDiagram-v2
    direction LR
    ITI --> Ready: elapsed
    Ready --> Reward: enter
    Reward --> ITI: leave

Tip

There are several ways to implement ROI activation, so feel free to explore different ideas. Consider using either Crop, RoiActivity, or ContainsPoint as part of different strategies to implement the enter and leave events.