Dragonfly’s secret hunting technique – internal models
Insects, in general, are quite small and hence, have smaller brains when compared to the average mammal. Therefore, insect behavior should not be as complicated as mammalian behavior, right? Most insect behaviors seen in the wild are thus, assumed to be largely reflex driven, i.e., the insect nervous system elicits a stereotypic response based on the stimuli sensed. Consequently, prediction and planning, thought to be high level processes requiring large number of neurons, were considered absent in insects. This study on dragonflies questions the underlying assumptions of insect neuroscience by demonstrating that dragonflies are indeed capable of both prediction and planning.
Prediction and planning, the basis of most human games, require three types of internal models. For example, to successfully catch a Frisbee, you intuitively do the following: 1) Calculate approximately where the Frisbee will land 2) Compute how many steps it would take to reach the Frisbee’s final destination. 3) Track the Frisbee while you move and position yourself to intercept it. Each of these, which you do without much thought, are pretty complicated computations by themselves. The first step of predicting the Frisbee’s final destination requires you to have a complex physical model of how the Frisbee moves, incorporating initial velocity, gravity, and lift generated due to the spinning Frisbee. The second computation of planning your trajectory towards the Frisbee’s final destination takes in your current and final positions as inputs and generates the movements necessary to reach the final position. This sort of computation requires an inverse model. Finally, while moving towards the Frisbee’s final destination, you need to continually keep track of its movements. However, while moving, from your perspective the Frisbee moves in the opposite direction. You need to subtract your self-motion, computed by forward models, from the Frisbee’s perceived motion, to determine its original motion. The combined output of the three models enables you to catch the Frisbee.
Dragonflies are superior aerial predators that hunt and capture small insects mid-air. The study looked at the neural basis of the dragonflies’ hunting prowess. Using an elaborate setup, the authors filmed dragonflies chasing prey using high speed cameras. They then computed the prey’s motion and compared it with the dragonflies’ response. By analyzing the data, the authors showed that dragonflies use all three types of models while hunting and capturing prey. It uses physical models to predict how the prey would move and inverse models to plan the best trajectory to intercept it. While tracking and following prey, it uses forward models to continuously subtract its motion from the prey’s motion to track it. Using all three models simultaneously, dragonflies capture prey mid-air, quite similar to how humans catch a Frisbee. No wonder dragonflies are such good aerial hunters! The authors now plan to investigate how such a small nervous system could execute such complex computations.
- Mischiati, M. et al. Internal models direct dragonfly interception steering. Nature 517, 333–338 (2014).