Optic flow stabilizes flight in ruby-throated hummingbirds

Flight control is crucial to ecologically relevant behaviors such as predator–prey interactions, courtship and foraging in dynamically and geometrically complex environments (Dudley, 2002a). Flight stability, the ability to resist and recover from perturbations, and flight maneuverability, the ability to change orientation and position (Guckenheimer and Holmes, 2013; Sefati et al., 2013; Dudley, 2002,b), are fundamental components of flight control. To actively stabilize flight by means of compensatory motor commands against external perturbations, such as turbulent air or wind gusts, detection of self-motion is required as a control input. Birds fly in complex three-dimensional environments and can detect self-motion using their visual, vestibular and proprioceptive systems (Benson, 1990). Additionally, filoplume-associated mechano-receptors can sense air flow (Gewecke and Woike, 1978). In cases of conflicting signals, evidence exists that the visual system overrules other sensory modalities for retinal image stabilization (Friedman, 1975; Gioanni, 1988). An important visual cue that indicates self-motion is optic flow: the resulting movement of the panorama, or visual surround, across the retinae (Gibson, 1958; Koenderink, 1986; Koenderink and Van Doorn, 1987). Optic flow arises from self-motion, whether voluntarily generated by the animal's motion or resulting from a perturbation, such as a wind gust during flight.

A variety of animals ranging from vertebrates to arthropods reduce rotation-induced optic flow by means of eye, head and/or body rotations (Walls, 1962; for reviews in insects, see Reichardt, 1969; Egelhaaf et al., 1988). These optomotor responses thereby stabilize the retinal image, facilitating the extraction of translational self-motion and depth information (Egelhaaf et al., 2012), and improving image resolution (Westheimer and McKee, 1975) as well as the ability to detect object motion (Nakayama, 1981).

It is known that birds use optic flow to guide various sensorimotor behaviors. Budgerigars choose flight paths that balance optic flow between the left and right sides, and regulate flight speed using cues based on optic flow, similar to insects (Bhagavatula et al., 2011; Schiffner and Srinivasan, 2015; Srinivasan et al., 1996; Baird et al., 2005; Fry et al., 2009). Optic flow parameters, such as tau, representing time-to-contact, are also used by hawks and pigeons to land (Davies and Green, 1990) and by hummingbirds to approach feeders (Lee et al., 1993, 1991). Recently, hummingbirds have also been shown to control hovering position using optic flow (Goller and Altshuler, 2014).

Additionally, certain wing and tail muscles in pigeons respond to head deflections induced by rotational visual stimulation during simulated flight (Bilo, 1992), indicating that optic flow is involved in rotational flight stabilization, as demonstrated in insects (e.g. Collett and Land, 1975; Lehrer and Srinivasan, 1992; Farina et al., 1995; Kern and Varju, 1998; Mronz and Lehmann, 2008; Theobald et al., 2009; Windsor et al., 2014), and suggested in hummingbirds and zebra finches (Srinivasan, 2001; Iwaniuk and Wylie, 2007; Eckmeier et al., 2013).

During locomotion, the optomotor response mostly separates the two components of optic flow that result from the bird's (1) translational motion as it moves from one place to another (translational optic flow) and (2) rotational motion as it changes from one orientation to another (rotational optic flow) (Egelhaaf et al., 2012; Eckmeier et al., 2008). Flight control requires both optic flow components because translational optic flow contains course, speed and depth information, while rotational optic flow informs a bird about its own rotations (Egelhaaf et al., 2012). Both translational and rotational components of optic flow are encoded in nuclei within the accessory optic system (AOS) of the avian visual system (Wylie and Frost, 1990).

Among birds, hummingbirds have an enlarged nucleus lentiformis mesencephali, one of the nuclei within the AOS involved in optic flow processing (Iwaniuk and Wylie, 2007; Simpson, 1984). Because of this neural specialization, combined with their stellar hovering and precision flight capabilities, hummingbirds are ideal species in which to test for a role of optic flow in avian flight stabilization (Iwaniuk and Wylie, 2007; Srinivasan, 2001; Greenewalt, 1960).

Here, we address whether hummingbirds use optic flow under free-flight conditions to control body orientation and flight position, in addition to stabilizing their vision. We present wide-field motion to free-flying hummingbirds in the form of rotations of their full 360 deg panorama by means of projecting moving images on a cylindrical surround (Fig. 1A,B, Movie 1). We test whether these visual perturbations elicit corrective responses in horizontal body rotations and flight paths. In general, wide-field motion should not evoke a robust and well-matched compensatory flight response if the visual surround is only used to stabilize vision. To solely stabilize vision, we would expect optokinetic head nystagmus without (or possibly with sporadic) optokinetic control of flight position or body orientation in the horizontal plane. In other words, we would expect that hummingbirds would only visually track their surround with smooth head rotations (until a fast head rotation returns the head to its original orientation), but without corresponding body rotations or changes in flight path (Gioanni, 1988). However, if the visual surround is also used to control body orientation and flight position, wide-field image motion should elicit corrective maneuvers during flight. We therefore hypothesize that hummingbirds perform horizontal body rotations and changes in flight position, in addition to making smooth head rotations. We expect these corrective maneuvers to reduce rotational and translational optic flow imposed by the rotating surround, indicating the importance of optic flow for flight control, as well as stabilization of the visual input.

Five female ruby-throated hummingbirds, Archilocus colubris, were trapped in Bedford, MA, USA, and studied in accordance with protocols approved by Harvard University's Institutional Animal Care and Use Committee.

The hummingbirds, hovering freely within a large vertical cylinder, were presented with a rotating visual pattern, covering their full 360 deg panorama (Fig. 1A,B, Movie 1). To generate a visual stimulus of surround rotations, four projectors (MW663, BenQ, Taipei, Xynyi, Taiwan) were distributed around a 1.2 m inner diameter, vertically oriented acrylic cylinder. The cylinder was coated for rear projection, extraneous visual cues were eliminated with black fabric, and an acrylic ceiling two-thirds of the cylinder height supported a syringe with artificial nectar. Thirty-two equal-width vertical bars (alternating black and white) were projected synchronously at 120 frames s⁻¹, with the image displacement between projection frames determining the rotation speed.

Given the short photoreceptor response time expected of hummingbirds (Healy et al., 2013), we projected the visual stimuli under low-light conditions to lower their critical fusion frequency. The projected images resulted in an illuminance of 38 lux at the center, and a Michelson contrast of 0.8 at the perimeter of the cylinder. The surround was either held stationary or rotated horizontally in either direction at 62, 98 or 134 deg s⁻¹, resulting in seven experimental conditions. Quantified movements and directions were expressed relative to stimulus direction, with positive values indicating the bird's movement with the surround. Each stimulation condition was initiated remotely by the researcher without knowledge of the bird's position within the cylinder, with the bird's response tracked for a period between 0.8 and 1.8 s.

Observed from the cylinder center, the spatial frequency of 0.04 cycles per degree combined with the rotation speeds (62, 98 and 134 deg s⁻¹) corresponds to temporal frequencies of 2.8, 4.4 and 6.0 Hz, which are within the range of the broadly tuned, fast directional neurons in the avian pretectum (Ibbotson and Price, 2001; Crowder et al., 2003) [e.g. (16 cycles/360 deg)×134 deg s⁻¹=6 Hz; Fig. 1A]. Under these free-flight conditions, the specific perceived frequencies vary depending on the bird's position within the cylinder and its perceptual focus. However, for the test here of a link between wide-field motion and control of body orientation and flight position, the specific spatial and temporal frequencies experienced by the bird are not critical, provided the frequencies remain within the range of the avian pretectum. The square-wave grating was kept constant and of a sufficiently low spatial frequency to avoid aliasing effects causing potential directional ambiguity at higher rotation speeds.

The hummingbirds flew freely within the cylinder and were recorded with two Photron 1024 PCI cameras (San Diego, CA, USA) at 500 Hz (Fig. 1A), for which sufficient illumination was provided by two 850 nm wavelength infra-red LED arrays that were imperceptible to the hummingbirds.

Using thermoplastic and cyanoacrylate adhesives, four white, 2–3 mm diameter, polystyrene markers were attached to each individual: on the head, two markers were positioned near the lateral ends of the coronal suture of the skull, and on the body, two markers were positioned dorsally over the spine, separated by 14 mm (Fig. 1B). Additionally, a 2 mm diameter dot of white non-toxic correction fluid was deposited on the bill. The maximum mass added to a bird was 0.04 g (1% body mass).

Using the two calibrated high-speed camera views, 3D positions of the markers were reconstructed in MATLAB (MathWorks, Natick, MA, USA; Hedrick, 2008) within a 0.2 m³ volume at the center of the cylinder, resulting in a spatial measuring error of less than 0.4 mm. The orientation in the horizontal plane of the marker-based direction vector was calculated for both head and body, with the time derivative giving the rotational velocity (Fig. 2, Fig. 3A–C). For the present study, ‘orientation’ refers to the horizontal component of orientation, which is more commonly used for 3D angular position. Both head and body orientation were referenced to the body orientation at time=0 s (start of surround rotation or start of stationary recordings). Therefore, the difference between the reported head and body angles represented the body–head offset throughout. Concentric flight velocity, i.e. the component of the flight velocity parallel to the nearest wall tangent, was based on a virtual head center marker (Fig. 1B, inset), with the time integral from the start of a recording giving the concentric flight distance (Fig. 1C–E).

Statistics were performed in JMP (SAS Institute, Cary, NC, USA), with the multiple least-squares linear regressions models corrected for individual effects (Fig. 3B–D).

The five hummingbirds hovered with a wingbeat frequency of 42.7±1.3 Hz. During trials in a stationary surround, voluntary fast head rotations, or head saccades, alternated with periods of no perceivable head rotation (stabilization; Fig. 1C). During trials with a rotating surround (±62, 98 and 134 deg s⁻¹), the hummingbirds displayed classic optokinetic head nystagmus, where fast head saccades alternated with slow phases of head rotations (Fig. 1D,E) (see Gioanni and Sansonetti, 1999). During the slow phases, the head always rotated in the same direction as the surround (Fig. 1C–E, Fig. 2). Furthermore, after approximately 0.5 s following stimulus onset (first observed displacement of the surround), the speed of the slow phases leveled off and closely matched that of the surround (Fig. 3A). To perform balanced comparisons of the behavioral responses across trials and individuals, we selected a 1 s period for all trials. Between 0.5 and 1.5 s after stimulus onset, head rotation velocity correlated with the surround speed (adjusted R²=0.98, P<0.001; multiple least-squares linear regressions model, which included the trends of individual birds as random effects, in order to correct for individual differences for all regressions; Fig. 3B).

For 81 head saccades observed in the five individuals, 75±18% (mean±s.d.) of these fast head rotations turned with the direction of the projected surround and 25±18% of the cases against (two-sample t-test, P<0.05; e.g. Fig. 1E).

Body rotations were more continuous, but generally followed head rotations (Fig. 1C–E). Surround speed during the period 0.5 to 1.5 s after stimulus onset predicted body rotation velocity (adjusted R²=0.17, P<0.001; Fig. 3C), though not as strongly as the slow phases of head rotation velocity (adjusted R²=0.98, P<0.001; Fig. 3B). When both slow and fast phases of head rotation velocity were included (and temporally corrected for a cross-correlation lag of −20±2 ms), horizontal body rotation velocity correlated with the surround more strongly (adjusted R²=0.40, P<0.001). Concentric flight velocity corrected for the bird's position relative to the cylinder center followed similar trends and correlated with the surround speed (adjusted R²=0.53, P<0.001; Fig. 3D).

Variability among individuals in all three variables (head and body rotation velocity, and concentric flight velocity) as a function of surround speed was consistently less than the variability observed within individuals (Fig. 3B–D, Table 1).

After stimulus onset, a latency of 52 to 62 ms was observed before the head began tracking the motion of the surround. This latency was conservatively estimated from the instant when mean head orientation exceeded 1 s.d. (52 ms) of steady head orientation to the instant when the mean exceeded 2 s.d. (62 ms; Fig. 4).

When presented with a virtually rotating cylindrical surround to induce optic flow, freely flying hummingbirds respond by rotating and flying with the surround (Movie 1; Fig. 2, Fig. 3B–D). Because a perturbation that rotates the bird to the left induces optic flow to the right, the response of tracking the surround with body and head rotations and following the surround with translational flight functionally serves to counter the perturbation caused by the visual presentation. Because the visual presentations do not result in vestibular sensations, this finding strongly indicates that hummingbirds rely on visual cues derived from motion of their surround to stabilize hovering flight.

The hummingbirds consistently and closely matched the imposed rotational speed and flew with the rotating surround, despite the differing spatial and temporal frequencies that varied depending on the birds' positions within the cylinder when presented with the rotating visual stimulus (Fig. 2, Fig. 3B–D). The tracking and following responses were robust, as illustrated by an additional 5 s trial, filmed at a lower frame rate, during which the bird responded for more than two full revolutions, partially flying backwards, and only reverted to normal hovering after the stimulus stopped (Movie 2). These findings indicate that under free-flight conditions, hummingbirds compensate for both rotational and translational optic flow components. Therefore, both components likely serve to control and stabilize flight maneuvers.

Previous findings that flight position in hummingbirds is controlled by optic flow (Goller and Altshuler, 2014) and that hawkmoths stabilize their flight position visually and possess visual interneurons sensitive to translational optic flow (Kern and Varju, 1998; Kern, 1998) corroborate our findings. However, our data cannot exclude the possibility that edge fixation on the retina, based on a vertical bar of the virtual surround employed in our experiments, may provide a second possible visual mechanism to control position.

Even though the accuracy (measured as tracking gain; Table 1) and variability (Fig. 3B,C) of the hummingbirds' responses to the imposed wide-field motion of a square-wave grating differed between head rotations and body rotations, head and body tracking of the surround was robust (see, for example, Movie 2). During slow rotation phases, the head tracked the surround more accurately (tracking gain=1.01) compared with the body (tracking gain=1.48), with perfect tracking signified by a gain of 1 (measured velocity/imposed velocity). This over-rotation of the body relative to the surround likely reflects the tendencies of head saccades to be directed with the surround and the tendency of the body to follow the head (Fig. 1). This tendency of the head to saccade with the stimulus motion, and thus against the perceived perturbation, is a distinguishing feature compared with the classical optomotor response, where the saccades (‘flick back’) are predominantly against the stimulus motion (see Türke et al., 1996). Notably, because birds tend to enhance gaze stabilization periods through brief, small eye motions relative to the head (Gioanni, 1988), the hummingbirds' eyes can be expected to track movement of the surround with even higher fidelity than the head. However, because the degree of eye movements relative to the head during flight in hummingbirds is unknown, we could not estimate the discrepancy between head movement and gaze.

The greater variability in tracking of the surround by body rotations than by slow phases of head rotations (Table 1, Fig. 3B,C) likely has multiple causes. First, body rotations are not saccadic in nature, especially when compared with head rotations (Fig. 1C–E). Consequently, the tendency of the body to follow the head saccades will increase variability in tracking the surround. Second, direct action of neck muscles provides greater control of head movement, whereas movements of the much larger body depend on aerodynamic forces generated by the wings. Body rotations, as observed here, are also likely needed to change flight velocity in order to follow the surround (Ros et al., 2011), in addition to tracking the surround through head and body rotations. Lastly, brief spontaneous maneuvers are characteristic of normal hummingbird behavior (Greenewalt, 1960), and may result in a superposition of normal hovering body rotations on top of the corrective body rotations tracking the surround. Therefore, more variable body stabilization compared with the head can be expected relative to the rotating surround used here to elicit visual perturbations.

We base our conclusions regarding optic flow and flight stabilization on the flight behavior of the hummingbirds as they tracked and followed the surround (Fig. 3). Interpreting underlying components of the observed ∼0.5 s latency between the stimulus onset and the birds' stabilized response to rotate and fly with the surround would require knowledge of eye motion and perceptual focus, which are challenging to achieve during free flight and generally unknown for birds (Fig. 3A) (see Gioanni, 1988). We can, however, estimate the delay between stimulus onset and the initiation of head rotations to be 52 to 62 ms (Fig. 4). The eyes likely respond earlier than the measured head movements (Gioanni, 1988). Nevertheless, such a delay is longer than visuomotor delays observed in insects (∼20 ms; e.g. Collett and Land, 1975), consistent with the smaller size and higher wingbeat frequencies of insects (Healy et al., 2013).

In conclusion, as experts in precise and highly maneuverable flight (Lee et al., 1991), freely flying hummingbirds appear to use similar sensory mechanisms to stabilize flight as insects. Our findings corroborate recent findings that Anna's hummingbirds control hovering position in response to projected visual motion (Goller and Altshuler, 2014). Our currently reported visual control of horizontal rotation and flight position in ruby-throated hummingbirds further supports the general view that birds and insects rely on both rotational and translational optic flow components to control flight maneuvers. This suggests convergent evolution on similar principles for robust visually guided flight performance in both groups of flying animals.

We thank P. A. Ramirez for care of the animals, and are grateful to Y. Hong, A. N. Ahn, A. M. Mountcastle, C. D. Williams and S. Ravi for helpful contributions.