Responsive robotic prey reveal how predators adapt to predictability in escape tactics

To increase their chances of survival, prey often behave unpredictably when escaping from predators. However, the response of predators to, and hence the effectiveness of, such tactics is unknown. We programmed interactive prey to flee from an approaching fish predator (the blue acara, Andinoacara pulcher) using real-time computer vision and two-wheeled robots that controlled the prey’s movements via magnets. This allowed us to manipulate the prey’s initial escape direction and how predictable it was between successive trials with the same individual predator. When repeatedly exposed to predictable prey, the predators adjusted their behavior before the prey even began to escape: prey programmed to escape directly away were approached more rapidly than prey escaping at an acute angle. These faster approach speeds compensated for a longer time needed to capture such prey during the subsequent pursuit phase. By contrast, when attacking unpredictable prey, the predators adopted intermediate approach speeds and were not sensitive to the prey’s escape angle but instead showed greater acceleration during the pursuit. Collectively, these behavioral responses resulted in the prey’s predictability having no net effect on the time taken to capture prey, suggesting that unpredictable escape behavior may be advantageous to prey in fewer circumstances than originally thought. Rather than minimizing capture times, the predators in our study appear to instead adjust their behavior to maintain an adequate level of performance during prey capture.

Rapid evasive responses are a vital tool prey use to minimize capture by predators (1, 2). Despite their ubiquity, it can be challenging to demonstrate the benefit of escape strategies, due to the difficulties involved in designing studies which integrate realistic predation with manipulation of prey behavior that experimentally controls for confounding effects. Studying the behavior of real predators is crucial when attempting to demonstrate the adaptive value of prey adaptations, especially when these are dependent on features of predator cognition (3–5). This applies particularly to unpredictable escape behavior by prey, which is thought to enhance prey survival by compromising the ability of predators to anticipate the movement of their target (6). Although unpredictable escape tactics are widespread taxonomically (7, 8), we know little about how real predators respond to unpredictability in prey escape strategies and whether this prevents predators from adjusting their behavior over multiple interactions with prey (9, 10).Controlled experiments in which human predators target continuously moving virtual prey have demonstrated that abrupt and unpredictable changes in direction reduce the accuracy of prey targeting (11, 12). However, it is unknown whether the survival advantage conferred by unpredictable motion also applies against nonhuman predators. Additionally, the escape responses of prey which are initially stationary are common in nature, as numerous prey taxa freeze once they have detected a potential threat or remain motionless to avoid detection by predators, before eventually fleeing only once a predator gets too close (1, 13–15). One way for stationary prey to be unpredictable is to vary the initial escape angle from one encounter to the next (16). This is a distinct tactic to the unpredictable movements made by prey which move continuously regardless of the presence of a predatory threat (6) or the abrupt turns made by some prey in anticipation of a predator’s attack (17). Although theoretical models predict that for a predator of a given speed, prey should select a single optimal escape trajectory which maximizes the distance from an approaching predator (18, 19), predators might anticipate the movements of prey which repeatedly escape at a fixed angle relative to their approach (20). Contrary to expectations based on a single optimal escape path, a wide range of prey species show a substantial degree of variability in their initial escape angles (16), including amphibians (21), crustaceans (22, 23), fish (24–27), insects (28, 29), mammals (30), and reptiles (31). Given that this variability is so widespread taxonomically, investigating whether it represents an antipredator strategy aimed at generating unpredictability could have major implications for our understanding of prey escape behavior (32, 33).Many predator-prey interactions are typified by feedback between the attacker and the target (34), making it difficult to disentangle the effects of prey defenses on predators from the impact of predator behavior on prey using a purely observational approach. One way to determine the importance of prey defensive tactics is to present real predators with standardized virtual prey, whose movements and behavior can be precisely controlled and experimentally manipulated (35–39). However, previous experiments with virtual prey have used unresponsive prey items which do not react to predators, and do not allow the predator to capture prey and be rewarded, making it extremely challenging to study repeated interactions between predators and prey. These limitations can be overcome by using interactive robotic prey (40).To study the effect of unpredictability in prey escape on predators, we developed an experimental system [Fig. 1A; see also Swain et al. (41)], in which artificial robot-controlled prey were programmed to flee from blue acara cichlid (Andinoacara pulcher) predatory fish. Blue acaras are opportunistic predators with a broad diet but actively pursue highly evasive prey such as Trinidadian guppies (Poecilia reticulata) (42, 43). Prey initiated their escape response once the predator had approached within a threshold distance (Fig. 1B), mimicking the tendency of many prey to flee from a distant predator at submaximal speeds (14, 44). After an initial period in which groups of blue acaras were trained to attack the prey (the training period, SI Appendix, SI Methods), individual predators were repeatedly presented with prey which escaped either in predictable or unpredictable directions over 20 successive experimental trials (the test period). For individuals assigned to the predictable treatment (which acted as the control), prey escaped at the same angle relative to the predator’s approach from one trial to the next, whereas in the unpredictable treatment, prey were programmed to flee in random directions over successive trials (Fig. 1C). To successfully capture prey, pursuit predators must respond to changes in prey direction, which occur at the start of a chase (45–47). Across trials with predictable prey, the predators had the opportunity to gain reliable information about the prey’s likely escape direction, in contrast to the unpredictable treatment where the prey’s escape angle in previous trials was not a reliable indicator of its escape direction in future encounters. If unpredictable escape behavior is adaptive, increased uncertainty about the prey’s likely escape direction in the unpredictable treatment should reduce the performance of the predator in these trials, with slower speeds of approach (i.e., before the prey respond), longer times taken to capture prey, and/or greater kinematic costs resulting from higher speeds, increased acceleration, and more turning during the pursuit.Open in a separate windowFig. 1.The robotic prey system. (A) Diagram (not to scale) showing a side view of the experimental system, with the Bluetooth-controlled robot situated on a platform underneath the experimental tank and the webcam used to monitor the predator’s movements suspended overhead. The prey’s movements are controlled by the robot via magnets, enabling the prey to escape from an attacking predator once the predator approaches within 27 cm of the prey’s starting position. See ref. 41 for a similar system designed for robotic predators attacking prey fish shoals. (B) Prey escape angles were defined relative to the predator’s approach direction. (C) In the predictable treatment, prey escaped at the same initial angle over successive trials (the escape angle varied between individual predators). In the unpredictable treatment, the prey’s initial escape angle varied randomly from trial to trial. While the experiment manipulated the prey’s initial escape angle, the prey’s subsequent escape trajectory was fixed as a straight-line path in both treatments.