The project was prompted by the observation that spider webs are highly complex structures leading the team to speculate that rather than simply serving as traps or snares they are in fact used for a variety of more sophisticated purposes. For example, a spider might use its web as a signal processor to locate and categorise events like different prey or mates – as well as locating when and where there are broken threads.
“Spiders also probe their webs by exciting them and then ‘listening’ to the echo with highly sensitive mechano-receptors called slit sensilla or lyriform,” says project co-leader, Dr Helmut Hauser, a lecturer in robotics at the University of Bristol.
According to Hauser, this particular use relates closely to the concept of morphological computation, commonly used in robotics, which is based on the realisation that the physical bodies of biological systems – like animals, plants and cellular structures – play a ‘crucial role’ in intelligent behaviour. This means that forms and structures of organisms – or their morphologies – are such that they are capable of carrying out functions that can be characterised as computational, without the need to use a brain.
“One could also say that in nature ‘computation’ does not happen just in the brain, but is partly outsourced all over the body,” he says. “It is important to note that conventional robotics does not consider this at all. Robots are designed under the assumption that first there is a body and then it is controlled centrally from a brain.”
Although this approach works fine in controlled and predictable environments like factory assembly lines, Hauser argues that it ‘fails completely’ when applied to complex, unpredictable and dynamic environments like living and working spaces – a major reason why we still haven’t seen robots sophisticated enough to carry out household chores, despite the fact they’ve been promised for around 60 years.
Robotic prototypes
In an effort to address such fundamental challenges, Hauser says the project team will develop a much deeper understanding of exactly what’s going on in spider webs with respect to computation. In doing so, it will clarify the underlying design principles relating to the intelligent morphologies (or structures and forms) capable of carrying out useful computations through vibration, in an effort to develop what Hauser calls a ‘fundamental general model’ that goes beyond spiders’ webs. Using these principles, the team will then design and build novel vibration and flow sensors – as well as spider-like robotic prototypes capable of deploying these sensors when required.
“Our partners in Oxford will develop and extend experimental setups to systematically investigate the computational capabilities of real, naturally spun webs,” says Hauser. “Based on these findings, we will develop theoretical models and a simulation framework to investigate and generalise the idea to go beyond web morphologies and materials other than silk.”
Hauser reveals that the Bristol team will also build the real sensor prototypes for flow and vibration sensing – which ‘will not look like spider webs, but will more based on the simulation results’ – and assume responsibility for building robots that are able to deploy the sensors on demand. Throughout the entire process, he stresses, there will also be a ‘close interaction between the biological, the theoretical and the prototyping parts of the project’.
“The idea is based on the previously described concept of morphological computation,” he says. “Computational aspects like filtering, integration, nonlinear combination and fading memory will be directly carried out in these morphological structures that are in the mechanical structure. As a result, the readout and the additional digital overhead will be minimal in complexity, resulting in highly robust systems that are fast in computing and fast in learning.
“Basically what we do is outsource computation to the body itself. We have demonstrated this concept previously in simulations, as well as on a number of robotic prototypes, for example on a silicon based octopus. However, this project goes beyond that by looking into finding a general framework and how this can be used in a real-world application, specifically, in sensing.”
In coming years, Hauser predicts this sensing technology will be well suited for use in a wide range of design and engineering applications, including where observations of vibration or flow patterns are crucial in detecting anomalies – such as on wind turbines, airflow through tubes or where vibrations act on big buildings, structures or machines.
Hauser concludes: “Our sensors are expected to be especially robust and potentially adaptive and highly autonomous. For example, they could learn normal vibration patterns on their own and set off an alarm if they detect anomalies.
Morphological computation One way of grasping the concept of morphological computation is to consider the example of the human running style. As Hauser explains, when we run the muscle-tendon systems in our legs act like springs that are able to react and adapt to uneven ground, resulting in ‘a highly stable movement without the need for communication with the brain whatsoever’. Another example is the Erodium seed, which drops to the floor but doesn’t do anything until the surrounding environment is wet or humid enough – triggering the unravelling of a spiral and a corresponding drilling movement. “The seed literally plants itself into the ground when the conditions are favourable. That is intelligent behaviour without a brain,” says Hauser. “There are many more examples all over nature ranging from clever control to intelligent sensing. In our research, we try to formalise these ideas in mathematical frameworks, but more importantly we build robots based on these ideas. The resulting machines are more robust and often highly adaptive and energy efficient.” |
