Learning to escape robotic spiders: predator

avoidance learning in bumblebees

Tom Ings & Lars Chittka

Outline

• Challenges facing bees

– Finding food BUT avoiding predators

• Colour change in spiders

• Impact of crab spiders on bees

• Dynamics of predator avoidance

– Avoidance learning paradigm

– Predator crypsis, learning & memory

Finding food

• Complex landscapes – too much choice!

Avoiding predators - invertebratesSit-and-wait ambush predators

Misumena vatia – colour change

Yellow White

Aims

• Does predator crypsis disrupt avoidance learning?

• Quantify dynamics of predator avoidance learning – bumblebees

• Does the foraging behaviour of bees alter in response to predation risk?

Time

Avoid

ance Attack

Effects of crab spiders on bees & plants

• Consumptive – reduce density & visitation

– Most (>90%) attacks unsuccessful (ample opportunity to learn)

• Non-consumptive – FEAR

– Avoid flowers with dead bees & spiders

– Avoid flower patches with high densities of spiders (sometimes)

• Lower seed set when spiders present

e.g. Dukas (2001), Gonçalves-Souza et al (2008)

Avoidance learning paradigm

Meadow of artificial flowers

10 cm

100 cm

72 cm

72 cm

Avoidance learning paradigm

“Safe” flowers

(trap inactive)

Trapping mechanism

“Dangerous”

flowers(active trap)

+

Life-sized 3D

spider model

“Robotic” spiders

Feeding

mechanism

Remotely

controlled

solenoid

Simulated predation attempts

Simulated predation attempts

2) Trap closed by

remote control

1) Bee lands to feed

3) Bee held for 2s

1) Learning to avoid predators in a single flower species meadow

vs

Questions

• Does spider crypsis affect avoidance learning & memory?

Hypotheses

• Bees are less accurate when spiders are cryptic

• Avoidance of spiders is well maintained

Design: avoidance learning

2 treatment groups

conspicuous

(16 bees)

cryptic

(16 bees)

Ings & Chittka (2008) Current Biology, 18

Colour vision in bees

• 3 colour receptors

Adapted from: Chittka & Brockmann (2005) PLoS Biology, 3

Wavelength (nm)

UV Blue Green

Rela

tive

sensitiv

ity

Colour differences: bee colour space

0.03 units

Blue

GreenUV

0.26 units

Bees can discriminate

0.1 units

Spider

Flower

Design: avoidance learning

• 25% chance of attack

• Learning curves

• Flight behaviour (3D video tracking)

2 treatment groups

conspicuous

(16 bees)

cryptic

(16 bees)

Ings & Chittka (2008) Current Biology, 18

Procedure

• pre-training (no spiders)

• learning (4 spiders & attacks)

• mid-term memory (spiders, no attacks)

• reinforcement (as training)

• overnight memory (spiders, no attacks)

0.00

0.05

0.10

0.15

0.20

0.25

25 50 75 100 125 150 175 200

Flower visit

Pro

po

rtio

n d

an

ge

rous f

low

ers

Accuracy: probability of attack

RandomLearning

0.00

0.05

0.10

0.15

0.20

0.25

25 50 75 100 125 150 175 200

Flower visit

Pro

po

rtio

n d

an

ge

rous f

low

ers

MM OR

Memory

Accuracy: probability of attack

Learning

Flight data: rejection of flowers

• Bee enters flower

zone

• Inspects flower

• Does not land &

feed

• Time in flower zone

Flower zone

Feeding zone

Flower zone

Feeding zone

0.4

0.5

0.6

0.7

0.8

0.9

1.0

1.1

1.2

25 50 75 100 125 150 175 200

Visit number

Du

ratio

n (

s)

Speed: rejection of dangerous flowers

Learning

MM OR

0.4

0.5

0.6

0.7

0.8

0.9

1.0

1.1

1.2

25 50 75 100 125 150 175 200

Visit number

Du

ratio

n (

s)

Speed: rejection of dangerous flowers

Learning Memory

‘False alarms’

P = 0.04

LearningN

um

be

r of

safe

flo

we

rs r

eje

cte

d MM(~15mins)

OM(~24hrs)

Summary: speed-accuracy trade-off

• Bees rapidly learn to avoid conspicuous & cryptic spiders

• Bees do not make more mistakes when spiders are cryptic

• Rejecting dangerous flowers takes longer if spiders are cryptic

• Bees reject more safe flowers when spiders are cryptic

• Reduced foraging efficiency

• Reduced fitness

• Cryptic spiders have no advantage - why use camouflage?

• How does flight behaviour change in response to spiders?

Conclusions