Intention Estimation Using Set of Reference Trajectories as Behaviour Model

Abstract

Autonomous robotic systems operating in the vicinity of other agents, such as humans, manually driven vehicles and other robots, can model the behaviour and estimate intentions of the other agents to enhance efficiency of their operation, while preserving safety. We propose a data-driven approach to model the behaviour of other agents, which is based on a set of trajectories navigated by other agents. Then, to evaluate the proposed behaviour modelling approach, we propose and compare two methods for agent intention estimation based on: (i) particle filtering; and (ii) decision trees. The proposed methods were validated using three datasets that consist of real-world bicycle and car trajectories in two different scenarios, at a roundabout and at a t-junction with a pedestrian crossing. The results validate the utility of the data-driven behaviour model, and show that decision-tree based intention estimation works better on a binary-class problem, whereas the particle-filter based technique performs better on a multi-class problem, such as the roundabout, where the method yielded an average gain of 14.88 m for correct intention estimation locations compared to the decision-tree based method.

Predicting Agent Behaviour and State for Applications in a Roundabout-Scenario Autonomous Driving

“…The model therefore is less generalisable compared to the geometric model presented above, as the trajectories recorded inside a single specific roundabout would represent vehicles’ behaviour only in that roundabout. This modelling approach was also used in our previous study [39]. A set of reference trajectories constituting such a roundabout model is presented in Figure 1b.…”

“…Having one of the above three models at hand, as a new query trajectory begins to comes in, the first task is to predict what exit direction is the agent expected to take (to leave the roundabout) so that more precise predictions such as future agent state can be estimated. For the purpose of exit direction classification, we use and build upon our previous work [39], where we proposed exit direction classification using a decision-tree based as well as a particle-filtering based approach, using heading angle, lateral position offset and speed as feature attributes. The study showed that, in a multi-class problem such as the roundabout, the particle-filtering based approach performed better.…”

“…The work presented in [39] proposes exit-direction classification based on a set of reference trajectories and hence is comparable to the model presented in Section 3.1.2 above. This means that the number n of reference trajectories in the model is not fixed and can change depending on factors such as availability of data and complexity of problem at hand, and can usually range between a few tens to a few hundreds [39].…”

“…The work presented in [39] proposes exit-direction classification based on a set of reference trajectories and hence is comparable to the model presented in Section 3.1.2 above. This means that the number n of reference trajectories in the model is not fixed and can change depending on factors such as availability of data and complexity of problem at hand, and can usually range between a few tens to a few hundreds [39]. The fundamental difference, in terms of exit direction classification, when compared to the other two models presented in Section 3.1.1 and Section 3.1.3 is that in these two cases the reference trajectories in the model are fixed to one trajectory per class, i.e., a total of 16 trajectories in each model for the case of roundabout.…”

Intention Based Comparative Analysis of Human-Robot Interaction

“…The presented approaches [6], [7], [8] focused warehouse scenario, behavior modeling using trajectories, and collision avoidance considering the human intention to avoid a collision, to model the human behavior for navigation and successful navigation in warehouse scenario. Once again the prior definition of human intention was not defined.…”