the human input should result in a collaborative effort between human and robot. Although human input can be explicit and HRI can produce cooperative behaviour through direct interaction, it can also be implicit and behave considerately through passive observation. This latter notion of a robot’s considerate behaviour is the seed to the motivation of this thesis, by using passive input from observing pedestrians move.
1.1
Motivation
As observed through the evolving HRI field, robot interfaces are becoming more human-like. Already humanoid robots are designed to walk like humans [26, 27], talk like humans [34], even used to assist humans as they walk [35], and a robot should be able to interact with a human as easily as another human would. Robots that are achieving more collaborative interactions is developing the field social robots, including how they move around with us. Designing a human-aware robot navigation that can be implicitly understood by humans will increase the diversity of HRI implementations.
HRI with autonomous mobile robots (AMRs) began in the previous century, gaining notability with museum tour guides (e.g. [36, 37]). These implementations relied heavily on patience towards pedestrians, pausing the AMR’s movement un- til any pedestrians had moved out of the way. The AMRs navigated differently from huamns and did not implement a co-operative interaction strategy with the pedestrians in order to navigate, which is often how pedestrians themselves are ob- served to work together [38, 39]. Even in recent implementations, which intend to adapt more human-like behaviours (e.g [40–42]), the robots move at a much slower pace than the pedestrians they navigate around. Pedestrians manoeuvre around the robots, allowing the robots to move relatively unhindered along a straight path. Even human-human navigation is still a highly researched area (e.g [43–45]), pro- viding evidence that our knowledge of how people successfully navigate together is insufficient. To design a robot capable of successfully navigating among pedestrians, when even the navigation system of the pedestrians are unknown, presents a very difficult real-world challenge.
1.1.1 Improving Human-Robot Interaction
Creating navigation systems capable of negotiating though crowded pedestrianised environments can be beneficial to advancing all forms of autonomous robotics. It can progress HRI by reducing the gap between the environments that humans and robots can safely co-inhabit, as well as increasing navigational autonomy overall. One of the main reasons for mobile robots still being segregated from humans is
due to their inability to manoeuvre in similar ways. Even very slight adjustments in movement can affect the equilibrium required for multiple dynamic agents to successfully navigate together, such as with flocking algorithms [46], due to collective behaviour and swarm intelligence [47]. For example, cyclists are segregated from pedestrians due to their larger size, faster speed, and reduced ability to manoeuvre. Despite the fact they are both moving humans the behaviour of the human cyclist would not be able to navigate amongst human pedestrians because of these subtle differences. Robots are segregated even more due to additional factors, such as: slower movement speeds, no adaptability (a cyclist can always dismount and become a pedestrian), and dissimilar navigation protocols.
For multiple dynamic agents to interact within the same space successfully, each must be able to anticipate one another’s movements. If a pedestrian does not move within a crowd with the same behaviour as the others, they will become a problem for those they interact with. Their responses will be unexpected, which would in- crease confusion, as well as increasing collision potential and congestion [39]. Some current robots may be likeable to humans in regards to physical appearance, however they have not been integrated into co-habitation environments as they are unable to navigate within them effectively (e.g. [25]). Robots are unable to perfectly commu- nicate with humans, due to a lack of advanced intelligence and adaptive cognitive ability, associated with human reasoning. This means that each robot can only operate within specific parameters, designed for specific environments.
For robots to manoeuvre through crowds of pedestrians they do not require the ability to process information like a human1. However, they should at least appear
to adopt human-like qualities, so they are capable of easily integrating with mini- mal disruption. The research presented in this thesis does not propose a complete robot platform which is indistinguishable from a human, rather it provides a novel framework for a robot’s navigation system. The focus is on the area of pedestrian crowd navigation, using a novel path planner to predict paths in order to move considerately around pedestrians and avoid potential collisions before they have a chance of developing. This considerate navigation strategy (CNS) relies on an im- plicit response, Item a2.2, to a minimal amount of passively observed pedestrian input. Rather than requiring explicit command inputs from a user, the CNS will move an AMR so that it causes minimal disruption to any surrounding pedestrians. As well as provoking implicit cooperation between the AMR and other pedestrians, by manoeuvring in ways similar to other pedestrians. A specific localisation system is not considered within this thesis, as the focus is on planning considerate paths using a novel considerate path planner (CPP). The AMR’s ability to correctly de- termine its current position and goal location is a considerably large research area
1
The Human Brain Project demonstrated the complexity of the human brain’s processing power, with their simulation taking 20 minutes to emulate 1 second of brain activity [48]
1.1 Motivation 5 of its own, such as SLAM (simultaneous localisation and mapping), and the CPP is only designed to plan considerate paths between AMR and goal from whatever data is provided on their positions.
1.1.2 Real-World Applications
Applications of HRI with AMRs have already proved effective in assistance for reha- bilitation [49–51]. The development of this thesis’ CPP will improve these current applications. Especially for visually or physically impaired individuals, if imple- menting it into autonomous mobility devices that can assist how they move. The CNS can also be beneficial for emergency response services that require fast access through a crowd, in order to attend a medical emergency. As the CNS navigates with implicit consideration it will not rely on pedestrian cooperation, and so could efficiently navigate through a crowd faster and without hindrance.
Autonomous Mobility Devices
For individuals suffering from blindness, or visual impairment, navigating through a crowd can be difficult as it is not possible to assess how other pedestrians are moving in front of them. Although audio and touch are still available, audible details cannot define the locations or orientations of individuals whilst touch requires a direct contact that should ideally be avoided. Current non-robotic methods of navigation for the blind involve assistance from guide dogs, which are only capable of manoeuvring around obstacles if the owner already knows the route2. Robotic
devices, such as the UltraCane [52], only detect objects in the immediate vicinity and require the operator to decide on what action to take based on vibrations they receive through the handle. The problem with these systems is that they only assess immediate obstacles. The motivation for this thesis is to allow visually impaired pedestrians to be lead through much more complex dynamic environments. The CNS would not require any of the other crowd members to make allowances for it, and the user will be directed through the crowd without hindrance.
A further possibility includes the integration of the CNS into autonomous wheelchairs, to help benefit those with severe movement difficulties. For individuals that cannot successfully move their wheelchairs around without assistance, the CNS will allow the wheelchair to plot considerate paths that it can then move along by itself. Emergency Response
When attending to someone who requires first aid often time is of the essence, especially for those who have suffered a heart attack or stroke. If such an incident occurs in a crowded area, being able to get to the scene as efficiently and quickly
2
as possible can add valuable seconds required to saving someone’s life. As the CPP uses predictive models to anticipate crowd movements, the planned path could be more efficient than if performed by a human. This is not only beneficial in getting to someone as quickly as possible, but also moving them out and away from the crowd (e.g. to an ambulance), in order to get them the attention they require as soon as possible.