«James Walker 1st year DPhil, Interdisciplinary Bioscience Doctoral Training Partnership Supervised by Prof Fritz Vollrath and Dr Beth Mortimer Oxford ...»
Exploring robotic solutions in the control of
running after induced self-amputation in
Tegenaria domestica (Agelenidae)
1st year DPhil, Interdisciplinary Bioscience Doctoral Training Partnership
Supervised by Prof Fritz Vollrath and Dr Beth Mortimer
Oxford Silk Group, University of Oxford
5th January 2015 – 5th April 2015
The self-amputation of legs in spiders is common but is costly, requiring compensation strategies for
behavioural tasks. Many spiders will subsequently regenerate their legs but this leads to further control challenges as partially regenerated legs must be integrated into behavioural algorithms based on modified sensory input. Here, a series of induced self-amputation treatments, resulting in successive legs lost from the same side, were used to introduce bilateral asymmetry into Tegenaria domestica. The compensation strategies were studied by filming individuals running on a flat surface.
It was found that despite reductions in running speed and an increase in track curvature, these spiders have a remarkable ability to compensate after extreme leg loss. The removal of 3 legs from the same side had a dramatic effect on the timing and pattern of stepping of the remaining legs suggesting that gait control is not entirely reliant on preprogramed central commands. A stride-based analysis found that spiders responded to the loss of 3 legs by increasing stride length and duty factor for the remaining leg on that side while stepping frequency was found to be the most important parameter across all legs available to spiders in order to moderate speed and reduce track curvature. If validated with computation models, these compensation strategies could prove useful in the design of robust legged robots for applications in which limb loss is a potential issue. More generally, further interrogation of this system will elucidate the mechanisms of ‘morphological computing’ potentially guiding the design of embodied robotic controllers that outsource computation to their body. I conclude that the regeneration of legs in running spiders would provide an excellent system to further investigate mechanisms of compensation and the role of modified peripheral sensory information in control.
INTRODUCTIONThe self-amputation (autotomy) and subsequent regeneration of limbs is widespread in spiders (Vollrath, 1990; Roth & Roth 1984). It has been estimated that up to 70% of hunting spiders will lose a leg during their lifetime (Oppenheim, 1908), typically as a mechanism to evade capture by predators but also following intraspecific conflict. Despite the immediate benefits, limb loss inevitably brings challenges in the coordination of behaviours involving the integration of limbs, such as web building and running. This control problem is further complicated in spiders that regenerate their legs where instead of responding appropriately to a lack of sensory information they must respond to modified sensory information from partially regenerated legs that need to be integrated into behavioural tasks.
Given the high incidence of leg loss and regeneration, and the significant associated costs, it is reasonable to assume that spiders have preadapted compensation mechanisms to adjust behaviour when sensory input is modified.
Historically, the question of how animals control their movements has been a major concern of behavioural physiology with a recurring debate regarding the role of central vs peripheral control (Seyfarth, 1985). Whilst rhythmic behaviours can be maintained in the absence of patterned sensory input, there is increasing evidence that peripheral feedback by way of sensory receptors (such as proprioceptors) is essential for fine control of certain behavioural patterns and for reacting to the environment on demand (Barth, 2002). In spiders, support for peripheral filtering of sensory signals used to control behaviour has come from studies of receptor morphology. Lyriform organs (collections of strain receptors found near leg joints) on regenerated legs have an altered shape allowing them to modify signals sent to the CNS and potentially relieving the need for central compensations (Vollrath, 1995). This is an example of ‘morphological computing’ where some of the computational functionality can be embodied within the shape and material properties of the sensors rather than require that the CNS knows the exact position of each regenerated limb.
The coordination of regenerated limbs in spiders provides an excellent model system to investigate the neurophysiology of compensation and decentralised control that has not yet been explored. This approach to control, with an increased reliance on morphological computation, has important implications for the development of bio-inspired soft robots using materials that incorporate adaptive sensors and actuators therefore reducing the mechanical and algorithmic complexity involved in robot design (Pfeifer et al., 2014). Instead of suppressing the complex dynamics introduced by the compliant physical body, which is the reason why classical robots are built of rigid parts, the body can employed as a computational resource (Hauser et al., 2011). The general principles surrounding the control of legged locomotion in spiders have interesting applications beyond biological interest in the field of robotics. Legged locomotion has distinct advantages over the use of wheels in terms of energy efficiency and transversability over rough terrain (Wettergreen and Thorpe, 1992), however programming an appropriate and adaptable gait is a significant challenge within the field (Delcomyn, 1999). The development of robust compensation algorithms in the event of leg loss will be essential for certain robotic applications such as planetary surface exploration or for search and rescue operations (Trimmer, 2008) where legs may become trapped and need to be sacrificed.
Given the time constraints of this present study, and the seasonal constraints associated with spider regeneration, the focus here is on compensation after leg loss without regeneration. Previous studies have examined the effects of induced autotomy in spiders on a variety of parameters thought to be correlated with fitness such as running speed (Amaya et al., 2001, Apontes and Brown, 2005, Brown and Formanowicz, 2012), prey capture efficiency (Wrinn and Uetz, 2008, Steffenson et al., 2014) and development time (Wrinn and Uetz, 2007). All of these studies were able to detect some adverse effect of leg loss but none explored the mechanism used by spiders to compensate for the reduction in fitness associated with autotomy. The focus of this study is running performance in funnel-web spiders (Tegeneria domestica), an essential predator avoidance metric that will determine survival following autotomy. It is assumed that the control of running has been finely tuned by evolution to allow the most efficient path to be taken even in the event of leg loss in terms of both speed (Amaya et al., 2001) and track curvature. Previous observations have noted that when a spider loses one or two legs, it is still capable of comfortably moving around, through a combination of gait compensation and a reprogramming of the central pattern generators (Biancardi et al., 2011). Studies of insect locomotion indicate that feedback from peripheral sense organs is critical to the control of these compensation strategies when legs are removed as this results in a dramatic change in the stepping pattern of the remaining legs that is unlikely to be a preprogrammed compensation algorithm (Delcomyn, 1999) The key objectives of this pilot study are: a) to determine the suitability of this spider running system for future work on regeneration and morphological computing, b) to determine whether the underlying control of compensation strategies used by autotomised spiders relies on peripheral or central control, and b) to determine which gait parameters are available to spiders in order to compensate for leg loss. A series of autotomy treatments, resulting in successive legs lost from the same side, were used to explore the compensation strategies employed by spiders that were filmed running on a flat surface. Compensation was measured in terms of various gait parameters describing stepping dynamics from the literature on octopedal locomotion (Mayorga et al., 2013, Spagna et al., 2011).
MATERIALS AND METHODS
Collection of study animals I collected 39 funnel-web spiders, of the family Agelenidae, from urban environments in Oxford, UK (51°45'39.2"N 1°15'20.2"W) on 3 and 19 June 2015. All spiders were returned to the lab at the Department of Zoology (University of Oxford) and subsequently weighed, sexed, and classified.
Twenty-three spiders were of the genera Tegenaria domestica, (5 Males, 18 females) and 26 were identified as Coelotes terrestris (4 males, 22 females). Four of the spiders (10.3%) captured were found to already have a leg missing and were subsequently discarded. All spiders were housed in the laboratory in transparent boxes (25 x 16 x 12cm) allowing them to build funnel webs with a moist ball of paper to serve as a water source. Live fruit flies (Drosophila melanogaster) and blow flies (Calliphora vicina) were fed to the spider depending on their size.
Experimental procedure Five running trials were performed on 11 female, fully intact, T. domestica for 5 autotomy treatments (described below) resulting in a total of 275 trials. An additional 55 trials were conducted (5 for each spider) following the removal of all legs and roughly evenly spaced over 2 weeks in an attempt to detect improvement over time. For the duration of the experiments, spiders were maintained under controlled laboratory conditions at room temperature (19–23 °C), and a light cycle of roughly 13:11h light:dark.
Figure 1 – Track capture setup with spider positioned in the centre of board and camera setup to capture spider movement.
For each trial, the spider was carefully placed on a point marked in the centre of the board covered by an opaque beaker and allowed to recover for 3 minutes. The video camera was then started, the beaker was removed and the spider was stimulated to move with a small puff of air using a transfer pipette. Once the spider reached the edge of the board the camera was stopped and the spider was placed back in the centre of the board for subsequent trials again allowing 3 minutes to recover. At the end of each trial the board was cleaned in an attempt to remove chemical cues that could affect the track of subsequent trials. Trials were repeated for the same spider and leg treatment until 5 tracks had been recorded where the spider moved continuously without stopping and without abruptly altering direction.
Induced leg autotomy Five leg treatments were applied to each individual starting with intact spiders (8 legs) and then inducing the spider to autotomize one leg at a time from the same side until all legs from the same side had been removed (7, 6, 5 & 4 legs). The side from which autotomy was induced was randomly selected for each spider. Spiders were forced to autotomize their legs using a protocol described by Vollrath (1990) whereby the femur was held firmly by forceps leaving the animal free to self-amputate at the coxa-trochanter joint. This method resulted in the lowest loss of fluid and best recovery;
seeping had typically stopped within 5 minutes of leg autotomy and all spiders survived for at least 2 weeks after leg loss. Each leg loss treatment was separated by at least 20 hours allowing the spider to recover and the in-built autotomy to set. None of the spiders regenerated any legs, despite being immatures, two spiders having one post-amputative moult and Tegenaria being a genus known to regenerate (Mikulska et al., 1975).
Preliminary autotomy trials found that the different species collected responded differently to induced autotomy. Whilst spiders of the genus Tegenaria readily autotomised any leg held firmly at the femur, Coelotes terrestris demonstrated a far greater reluctance to lose a second or third leg and for no individual was it possible to induce the self-amputation of all 4 legs from the same side.
Data acquisition Video sequences were imported into MATLAB (ver. 2014b, Mathworks) allowing individual points to be tracked frame by frame using custom written software (Walker, 2009). For each sequence, the posterior tip of the abdomen was tracked for the entire length of the running bout resulting in 2D Cartesian co-ordinates of each point for each frame.
The track distance (x) was calculated for each track in MATLAB by summing the distance between adjacent points and converting into real space using a calibration.
Running speed (s) was calculated using the number of frames taken to travel the middle 50% of the track, in attempt to remove the acceleration at the start and deceleration at the end.
The curvature of each track (k) was calculated using a method described by Zollikofer (1994) using the inverse of the radius of curvature for 10 evenly spaced points along the track.
The tip of each leg was tracked when in contact with the board for sequences used to calculate gait parameters. Stepping pattern and gait parameters were averaged over 7 strides either side of the midpoint of the track (14 strides per track) in an attempt to avoid capturing abnormal gaits at the start and at the edge of the board at the end. Each gait parameter was calculated both for each leg in a track in order to determine differences in response to leg loss and averaged across all legs in order to identify overall adjustments in response to autotomy.
Stride length (l), defined as the distance between two successive footfall positions of the same leg, was calculated using the same calibration for distance described above Stepping frequency (f) was calculated as the inverse of the period of a stepping cycle, converted from the number of frames.
Duty factor (d), a key variable describing gait, was calculated as the ratio of duration of a foot contact interval to the stride duration.
The stride phase relationships were evaluated by means of the spatiotemporal contact sequences of support of the locomotive legs, see Figure 2a for an example.