Healthy adults take for granted the ability to navigate through complex and crowded environments. For adults with reduced physical or mental ability, particularly older adults, simple daily tasks, such as shopping or just socialising, can be extremely challenging. Normal environments can seem hostile and unfamiliar; other people can seem threatening, and crowds can cause feelings of claustrophobia. In general, leaving the familiar surroundings of home is stressful. As a result, those with reduced ability tend to avoid going out and suffer a consequent loss of physical and mental wellbeing, arising from reduced exercise, reduced fresh food and reduced socialising.
To address this problem, we have developed a portable motion planning device for crowded environments, to help reduce the stress of a user with impaired ability. Our device takes the responsibility of finding the best course of action in the local environment, freeing the user to focus on more important things, such as the physical effort of moving. We refer to this technology as the ‘local motion planner’.
The job of the local motion planner is to be aware of nearby pedestrians and obstacles, and to suggest a path that has the least chance of causing stress to the user. The motion planning problem we have addressed is particularly challenging because the environment is dynamic (people come and go arbitrarily) and we must anticipate future stressful situations (e.g., crowding). A merely reactive approach is not adequate, so our novel solution mixes reaction and prediction.
Our device observes the local environment in real time, employing the high performance visual sensing technology found in computer games consoles. This allows it to react rapidly to the changing environment. By recording the motion of other pedestrians, the device is able to construct a predictive mathematical model of their behaviour. Not all behaviour can be predicted in this way, so the model also includes an element of stochasticity (random behaviour). We use this model (the ‘social force model’) to simulate possible future scenarios, which we evaluate using ‘statistical model checking’ — a technology commonly used to quantify the performance of highly complex systems. The output is a suggested immediate action that will minimise the user’s stress and maximise progress.
We have verified the performance of our motion planner on low powered embedded computing hardware that is typical of portable devices. Our future efforts will be directed towards improving the predictive power of our mathematical model, by considering ‘behavioural templates’ (typical patterns of human interaction) and taking advantage of other visual cues (e.g., printed signs and facial expressions).