Virtual crowds are nowadays a requirement to create realistic digital worlds, as the beneficial effects of a lively background made of many moving characters in terms of visual quality and enjoyment is undeniable. However, because they are all animated by similar procedures, crowd characters often display a certain level of uniformity which can be detrimental for the plausibility of the virtual scenes. In this PhD, we are interested in exploring how to introduce variety in crowd simulations.
To animate such large amounts of characters in interactive applications, crowd engines are typically based on a two-step process: 1) the characters’ global 2D trajectories are generated by a crowd simulator, while avoiding collisions with other characters and the environment and walking towards given goals, and 2) an animation system animates the full body motion of each individual (often using motion capture data) on top of these global trajectories. While this two-step process is interesting for interactive applications for computational reasons, it also brings limitations in terms of variety of behaviours and motions. A certain level of uniformity can be visible in terms of global trajectory strategies, e.g. all characters usually display similar avoiding behaviours because they are all driven by a unique underlying simulation model. Uniformity can also be visible in terms of body motions as the animation engine is often only driven by the local position and velocity of the agents, without considering the presence of neighbours.
The aim of this PhD is to explore how perceptual variety can be introduced in crowd engines. Unlike previous works which focused on the number of distinct motions to use to create variety, as well as characters’ appearance [MLDSO08,MLHRC09], the goal here is to explore the possibilities of adding variety at two distinct steps in the crowd engine process.
The first objective will consist in exploring how to introduce variety at the simulation level, i.e., in global trajectories resulting from the crowd simulation. Crowd simulators usually compute the global displacement of hundreds of characters based on local interactions between simulated agents. Many models of varying complexity have been proposed in the literature (social forces [HM95], flocking rules [Reynolds87], velocity-obstacles [VLM08], synthetic vision [OPOD10], power law [KSG14], etc), each differing in terms of strategies used to drive characters and avoid collisions, but also in relation to their internal parameters. We therefore want to explore two complementary aspects: 1) how introducing variety in the parameters of simulation models will affect the visual quality of the crowd by creating variety in the global trajectories, and 2) how using different models in a single simulation will affect the overall visual quality by creating a range of distinct social behaviours.
The second objective is to further explore how to create variety at the animation level. Body animations are typically driven only by the position and velocity of the agents, without any consideration for the local interactions with neighbours. However, we demonstrated recently that driving secondary shoulder animations based on local interactions can improve the naturalness of crowd animations [HOKP16]. Still, much work needs to be done to understand how variations in body motions can be driven by other types of interactions (e.g., social interactions), and how such variations will affect the visual quality of crowd simulations.
Resuts and Impacts
To assess the effects of adding variety at the simulation or animation level in terms of visual quality, we will draw from methodologies in Psychophysics to conduct perceptual experiments and user studies. Typically, we expect to have identified the parameters of different simulation models that contribute the most to visual variety, and how to best visually combine these models. Similarly, we expect to better understand how body motions can be effectively driven based on local interactions to create more natural crowds. The objective is to transfer the results obtained in this PhD in interactive crowd simulations. Thus we expect a potentially wide impact of such results.
The candidate will work in the joined Inria / IRISA research centre located in Rennes. Inria (www.inria.fr) and IRISA (http://www.irisa.fr/) are amongst the leading research centres in Computer Sciences in France. The work will be supervised by members of the MimeTIC team, internationally recognised in the fields of Computer Graphics and Virtual Human Simulation, as well as by members of the Lagadic team, internationally recognised in the fields of Robotics and Computer Vision.
Requirements for candidacy:
- C/C++ recommended
- Basic knowledge of computer animation and computer graphics
- Interest in user experimentations
We are looking for motivated candidates, please send CV, a motivation letter, reference letters, and any relevant material to: firstname.lastname@example.org and email@example.com