Quantifying Learning in Young Infants: Tracking Leg Actions During a Discovery-learning Task

Podsumowanie

A method is described in which 3-4 month old infants learn a task by discovery and their leg movements are captured to quantify the learning process.

Streszczenie

Task-specific actions emerge from spontaneous movement during infancy. It has been proposed that task-specific actions emerge through a discovery-learning process. Here a method is described in which 3-4 month old infants learn a task by discovery and their leg movements are captured to quantify the learning process. This discovery-learning task uses an infant activated mobile that rotates and plays music based on specified leg action of infants. Supine infants activate the mobile by moving their feet vertically across a virtual threshold. This paradigm is unique in that as infants independently discover that their leg actions activate the mobile, the infants’ leg movements are tracked using a motion capture system allowing for the quantification of the learning process. Specifically, learning is quantified in terms of the duration of mobile activation, the position variance of the end effectors (feet) that activate the mobile, changes in hip-knee coordination patterns, and changes in hip and knee muscle torque. This information describes infant exploration and exploitation at the interplay of person and environmental constraints that support task-specific action. Subsequent research using this method can investigate how specific impairments of different populations of infants at risk for movement disorders influence the discovery-learning process for task-specific action.

Wprowadzenie

Task-specific actions emerge from spontaneous movements during infancy. It has been proposed that task-specific actions emerge through a discovery-learning process^1,2. Tasks are discovered by infants as they spontaneously move and explore actions which produce novel effects in the environment. Task-specific actions emerge as infants exploit the connections between their actions and their effects on the world around them. However, little is known about the precise processes that infants explore and exploit to learn to modify their spontaneous movements to perform task-specific actions. Here a method is described in which 3-4 month old infants learn a task by discovery and their leg movements are captured to quantify the learning process.

Figure 1: Infant kicking-activated mobile task. The center light-emitting diode (LED) attached to the rigid body of each foot (yellow circle) activates the mobile when it crosses the virtual threshold (red dashed line). Re-printed with permission from Sargent et al.³

This discovery-learning task uses an infant activated mobile that rotates and plays music based on the specified leg action of infants³. Infants placed supine under the mobile activate it by moving their feet vertically across a virtual threshold (Figure 1). This paradigm is unique in that as infants independently discover that their leg actions activate the mobile, the infants’ leg movements are tracked using a motion capture system allowing for the quantification of the learning process.

The experimental protocol includes two days of data collection. Day 1 consists of a 2 min baseline condition in which an infant kicks spontaneously but his leg actions cannot activate the infant mobile, followed by a 6 min acquisition condition in which the infant’s leg actions activate the infant mobile if the infant moves his feet vertically to cross a virtual threshold. This protocol allows for the quantification of infants’ spontaneous leg actions as well as the quantification of various aspects of the movements as infants explore the relation between their leg actions and activation of the infant mobile. On Day 2, in addition to the 2 min baseline condition and 6 min acquisition condition, a 2 min extinction condition is added in which the infant’s leg actions do not activate the infant mobile. This allows for the quantification of how infants change their leg actions when an already learned environmental response is discontinued.

In previous infant mobile paradigms, frequency of leg movement^4-6, specific hip and knee angles^7,8, or kicking a panel⁹ have been reinforced with mobile movement. Performance each day was defined as an increase in these leg actions during the acquisition or extinction condition as compared to the baseline condition^4-9. Learning across days was defined as an increase in these leg actions during the baseline or acquisition condition of Days 2 or 3 and the baseline condition of Day 1^5,6. These previous mobile paradigms demonstrate that infants increase the frequency of leg actions that are reinforced with mobile activation, however, they do not provide information on the movement options infants have available to them when learning the task. For example, if kicking rate is reinforced, infants demonstrate performance and learning when their kicking rate increases either when interacting with the mobile or when the mobile no longer activates. This demonstrates that infants can refine their kicking rate, but it is unknown if infants can refine their leg coordination pattern or torque production to generate leg actions that are not within their preferred movement repertoire.

This mobile paradigm is unique in that infants are required to demonstrate more refined leg action to activate the mobile than in previous mobile paradigms. In this mobile paradigm, the height of each foot above the table is computed during the 2 min baseline condition using position data from a light emitting diode (LED) attached to each foot. A virtual threshold is then set parallel to the table at a height that is within the upper range of the height of both feet during the baseline condition. During acquisition, the mobile rotates and plays music if either foot crosses the threshold. After 3 sec, the mobile stops and reactivates only if the infant moves the foot below the threshold, and then moves the foot vertically and again crosses the threshold. To activate the mobile for the greatest amount of time, infants need to move a foot above the threshold and maintain it against gravity for 3 sec, then quickly move the foot below the threshold and again move it above the threshold and hold it there for 3 sec, etc. This requires more refined leg action than simply increasing kicking rate.

Figure 2: Unfiltered position data of end effectors (feet) from a representative infant. Unfiltered position data from Day 2 of a 3 month old infant who demonstrated learning based on the individual learning criteria. The red line is position data of the z-coordinate of the light-emitting diode (LED) placed on the right foot. The blue line is position data from the LED on the left foot. Thick black line is the table. Dotted line is the virtual threshold placed 14 cm above the table as individually determined for each infant based on the height of their kicking during baseline condition of Day 1. X-axis is time labeled by 2 min intervals. Note how the infant moves his feet during baseline when the mobile does not activate and during the first 30 sec of acquisition 1, then he consistently keeps both feet off the table and moves his feet right around the threshold for the next 5½ min until the mobile no longer activates during the extinction condition.

The second unique feature of this mobile paradigm is that each infant’s leg action is tracked using state-of-the-art motion capture techniques to quantify how infants use their movement options to learn the task. Unfiltered position data of the LED on each foot that activates the mobile from one representative infant is included in Figure 2. Note how the infant moves his feet at various heights above the table during baseline and the first part of acquisition, but then moves both feet right around the threshold during the rest of the acquisition condition until the mobile no longer activates during extinction. This is one of many potential movement strategies to accomplish the discovery-learning task. The strategies can be quantified by computing three-dimensional kinematics and kinetics using position data acquired from the motion capture system. Specifically, the learning process is quantified in terms of the percentage of reinforced leg action (%RLA), which is equal to the duration of mobile activation, position variance of the end effectors (feet) which activate the mobile, hip-knee coordination patterns, and hip and knee joint torques.

Access restricted. Please log in or start a trial to view this content.

Protokół

The Institutional Review Board at the University of Southern California approved this study.

1. System Preparation

1. Set up the motion capture system. Please note: these steps are different for each motion capture system.
   1. Align the coordinate systems of the two motion capture sensors to that of one sensor by clicking “Perform New Registration” in the motion capture program, entering a collection time of 30 sec, clicking “Register,” and moving the registration object within the capture volume for 30 sec. When registration is successfully completed, observe a root mean square (RMS) registration error on the computer screen.
   2. Align the global coordinate system to the testing table using the registration object by clicking “Perform New Alignment” in the motion capture program.
      1. Define the origin by placing the registration object on the upper right corner of the testing table and clicking “Digitize” in the motion capture program. Define the Z-axis by placing the registration object on top of a box and clicking “Digitize”; the Z-axis is perpendicular to the table.
      2. Define the Z/Y+ plane by moving the registration object on the box along the length of the table and clicking “Digitize”; the Y-axis is parallel to the length of the table and the X-axis is parallel to the width of the table.
   3. Plug the LEDs into the two strobe ports and enter the number of LEDs per strobe port within the motion capture system program (24 for strobe port 1 and 20 for strobe port 2). Refer to Figure 3 for the number and location of each LED. Select missing data view to provide a strip chart-like display of LEDs being tracked in real time.
      Figure 3. Placement of LEDs. Strobe 1 consists of the rigid bodies (4 LEDs on each rigid body) for the right and left thigh, shank, and foot. Strobe 2 consists of the rigid bodies for the right and left pelvis, the sternum markers, and the individual LEDs.
2. Set up the infant mobile computer program.
   1. Input the number of minutes for each condition. For Day 1, input 2 for phase 1 (2 min baseline, non-reinforcement condition), 6 for phase 2 (6 min acquisition, reinforcement condition), and 0 for phase 3 (0 min extinction, non-reinforcement condition).
   2. For Day 2, input 2 for phase 1 (baseline), 6 for phase 2 (acquisition), 2 for phase 3 (extinction), and check “Use Zmin as Default” so that the threshold computed during baseline of Day1 will be the threshold used for the acquisition condition of Day 2.
   3. Choose “StreamframesAllFrames” and click “Send” to enable the mobile program to use data from the motion capture system to activate the infant mobile based on specified criteria.
3. Set up the video cameras.
   1. Initiate the video computer program for the three synchronized videos (right lateral, left lateral, overhead views).
   2. Start the additional video camera over the infant’s head to record facial expressions and eye gaze.

2. Infant Preparation

1. Describe the experiment to the parents and inform them to interact as little as possible with their infant.
   NOTE: Tell the parents that if the infant does not become fussy throughout the experiment, the parents should sit beside the infant outside of their view, however, if the infant becomes fussy there is a progression of interaction with the infant.
   1. First, ask the parent to say, “Everything is okay, I’m right here,” in a reassuring tone.
   2. Second, ask the parent to stand in the infant’s view while reassuring the infant.
   3. Third, ask the parent to either hold one of the infant’s hands or give the infant a pacifier.
      NOTE: The least amount of parent interaction necessary to keep the infant calm and alert is given and is ended as quickly as possible.
2. Undress the infant, place the infant under the infant mobile, and secure the infant to the table using a Velcro band placed across the trunk.
3. After the infant is secured to the table, place the sternum markers and pelvic, thigh, shank, and foot rigid bodies on the infant.

3. Infant Mobile Learning Task

1. Each day, initiate the mobile learning task by synchronously starting the motion capture system, mobile computer program, and video cameras.
   1. On both days from min 0 to 2, the baseline condition, observe the infant spontaneously kicking.
   2. On Day 1 during the 2 min baseline condition, observe as the infant mobile program continuously computes the threshold for mobile activation based on the z-data from one of the LEDs on the rigid body of each foot. Example, marker 9 on the right foot and marker 21 on the left foot. Marker 9 is the center LED on the right foot rigid body circled in yellow in Figure 1. Marker 21 is the center LED on the left foot rigid body.
   3. At the end of the 2 min baseline, the mobile program will set the threshold at a height of one standard deviation (SD) above the average height of both feet during the 2 min baseline condition.
   4. On both days from min 2 to 8, the acquisition condition, observe as the infant mobile rotates and plays music when the LED placed on either foot crosses the threshold computed during the 2 min baseline condition of Day 1.
      NOTE: Mobile activation will continue for as long as the foot is above the virtual threshold to a maximum of 3 sec. After 3 sec, the mobile will reactivate only if the infant moves the foot below the virtual threshold, and then moves the foot vertically and again crosses the threshold. This “3 sec rule” encourages active leg exploratory movements versus holding the feet above the threshold.
   5. On Day 2 from min 8-10, the extinction condition, observe as the infant kicks spontaneously without mobile reinforcement.
2. After the infant interacts with the mobile, collect a static calibration trial to define a local coordinate system for each leg segment and define a reference configuration for each body segment in space.
   1. Fix ten individual LEDs bilaterally to the infant’s skin using double-sided EKG collars at the following locations: lateral midline of the trunk below the tenth rib, greater trochanter of the hip, lateral knee joint line, ankle lateral malleolus, and distal end of the 5^th metatarsal.
   2. Hold the infant’s lower extremity in an extended, anatomical position for 5 sec. All joint angles in this calibration position are defined as 0°.
3. On Day 2, collect anthropometric data.
   1. Weigh each infant on a digital electric scale.
   2. Take the following measurements: total length of the infant; circumference at mid-segment of thigh, shank, and foot; width of knee (at the knee joint line), ankle (at the malleoli), and foot (at the metatarsal heads); and length of the thigh (greater trochanter to knee joint line), shank (knee joint line to lateral malleolus), and foot (medial malleolus to first metatarsophalangeal joint).

4. Data Analysis

1. Analyze performance and learning by computing the %RLA during each 2 min interval of the experiment using a custom computing language program such as Matlab. Compute the duration of time one or both of the LEDs on each foot that activated the mobile were above the threshold. Since the mobile does not activate after an interval of 3 sec, subtract the duration of time in which one or both LEDs were above the threshold for greater than a 3 sec interval.
   1. Measure performance of the group each day by statistically analyzing whether the %RLA during any one of the three, 2 min acquisition intervals significantly exceeds the 2 min baseline interval^3,4,7,9,10.
   2. Categorize individual infants as having performed the task each day if the %RLA during any one 2 min acquisition interval is equal to or greater than 1.5 times the %RLA in the 2 min baseline interval^3,4,6,9,10.
   3. Measure learning of the group across days by determining statistically whether the %RLA during the entire 6 min acquisition condition Day 2 exceeds the %RLA during the baseline condition Day 1^3,6.
   4. Categorize individual infants as Learners if the %RLA during the entire 6 min acquisition condition of Day 2 is equal to or greater than 1.5 times the baseline condition of Day 1^3,6,11.
2. Analyze arousal and attention by coding videotapes during each 2 min interval of the experiment. The arousal scale is defined as: drowsy = 1, alert and inactive = 2, alert and active = 3, fussy = 4, and crying = 5^3,8,11. The attention scale is defined as: 0 = not looking at the mobile, 1 = looking at the mobile^3,8.
3. Process position data and extract kicks using custom Matlab programs.
   1. Load position data files outputted from the motion capture system into a custom Matlab program to interpolate missing position data (maximum of 20 consecutive frames) using a cubic spline.
   2. Load the interpolated files into a custom Matlab program to (a) filter position data using a fourth-order Butterworth with a cut-off frequency of 5 Hz as determined from power spectrum analysis, and (b) compute the following joint angles: hip flexion/extension, hip abduction/adduction, hip external/internal rotation, knee flexion/extension, ankle inversion/eversion, ankle dorsiflexion/plantarflexion as described in ¹².
   3. Load the angle files into a custom Matlab program to extract kicks. Define the beginning of a kick as the onset of a continuous leg movement in which the hip or knee joint angle change exceeded 11.5° (0.2 radians) into either flexion or extension^3,9,13-15. Define the end of the kick as the frame of peak extension following a flexion movement or peak flexion following an extension movement^3,9,14.
4. For all kicks, compute kinematic parameters using custom Matlab programs.
   1. Compute position variance in the z-direction (vertical, task-specific direction) of the LED on each foot that activated the mobile³.
   2. Compute hip flexion/extension and knee flexion/extension joint correlations using Pearson correlation coefficients (r) at zero lag between hip and knee joint angle excursions. To compare correlations (r) among infants, convert hip-knee joint angle correlations to Fisher Z scores^3,9,15.
   3. Time-normalize the joint angle data, then compute hip flexion/extension and knee flexion/extension continuous relative phase (CRP) from the angular position/velocity data^16,17. Analyze the results of the CRP computation at the following five time points: (a) beginning of kick, (b) peak velocity of the first segment, (c) joint reversal, (d) peak velocity of the second segment, and (e) end of kick³.
5. For all kicks, identify non-contact kicks by viewing the synchronous video data. Compute kinetic parameters for non-contact kicks using custom Matlab programs.
   1. Compute the segmental mass and center-of-mass from equations modified for infants from Hatze’s anthropometric model for adults¹⁸. Compute the 3D moments of inertia of the thigh, shank, and foot segments from equations modified for infants from Jensen’s anthropometric model for adults¹⁹.
   2. Calculate the terms in the following equation of motion using the screw theory of spatial manipulations²⁰.
      
      is derived using the Lagrangian approach, where M(θ) is the inertia matrix,  the Coriolis and centrifugal torque matrix, N(θ) the gravitational (GRA) torques and T the muscle (MUS) torques.
   3. Compute joint torques using the 3D kinematic data from the non-contact kicks, body-segment inertial parameters, and the biomechanic equation of motion.
   4. Partition the net (NET) torque at each joint into motion-dependent (MDT), GRA, and MUS torque contributions²¹. NET torque is directly proportional to the accelerations at each joint. MDT torque is related to the passive torques associated with mechanical interactions among the moving interconnected segments of the limb. GRA torque is related to the passive force of gravity acting downward on the limb. MUS torque includes forces from active muscle contractions and passive deformations of muscles, tendons, ligaments, and other periarticular tissues.
   5. For the hip and knee separately, compute torque impulse as the magnitude of the contribution of each partitioned torque (MUS, GRA, MDT) to NET torque. Compute the positive or negative torque impulse (torque * time) during intervals in which the knee MUS torque acted in the same or opposite direction compared with the knee NET torque. Perform this same computation with knee GRA and MDT torques and hip MUS, GRA, and MDT torques. For the hip and knee separately, sum all positive and negative impulses for each torque component to yield a measure of the magnitude of the contribution of each partitioned torque impulse (MUS, GRA, MDT) to NET torque impulse.

Access restricted. Please log in or start a trial to view this content.

Wyniki

The learning process of young infants can be quantified in terms of the %RLA, position variance of the end effectors (feet), hip-knee angle correlation coefficients, and hip and knee joint torques. Each level of analysis provides unique information about how infants explore the relation between their leg actions and activation of the infant mobile during the discovery-learning process.

For the statistical analysis of %RLA and hip-knee angle correlation coefficients, mixed regression models wit...

Access restricted. Please log in or start a trial to view this content.

Dyskusje

Design of discovery-learning tasks for young infants

Discovery-learning tasks for young infants must be thoughtfully designed to assure that infants are independently discovering the contingency. In several mobile paradigms at the beginning of the acquisition condition, infants are either shown that the mobile activates by a non-contingent activation of the mobile^7,22 or the leg of each infant is passively moved by the investigator to introduce the infant to the contingency⁹

Access restricted. Please log in or start a trial to view this content.

Materiały

Name	Company	Catalog Number	Comments
Optotrak Certus Position Sensor, Far Focus, with stand	Northern Digital Inc	8800852
Optotrak Data Acquisition Unit II (ODAU II)	Northern Digital Inc	8800767
Optotrak Vinten Stand, Certus with Quick Fix Adapter	Northern Digital Inc	8800855.002
Certus S-Type, Standard Configuration	Northern Digital Inc	8800761
Marker (7 mm) pair, c/w RJII connector and 8 ft cable	Northern Digital Inc	8001029.001
AC Line Cord, Medical Grade, North America	Northern Digital Inc	7500010
Cubic Reference Emitter Kit - Certus	Northern Digital Inc	8800768
3 Pylon IEEE 1394 cameras	Basler	A6021c
Vixia HG10 camcorder	Canon	2183B001
Adhesive Disks	MVAP Medical Supplies	E401-500
Reversible head support	Eddie Bauer	52556
Softstrap Strap	Sammons Preston	A34960
Digital Pediatric Scale	Healthometer	Model 524KL