David Charypar

Particle-based fluid simulation for interactive applications

Realistically animated fluids can add substantial realism to interactive applications such as virtual surgery simulators or computer games. In this paper we propose an interactive method based on Smoothed Particle Hydrodynamics (SPH) to simulate fluids with free surfaces. The method is an extension of the SPH-based technique by Desbrun to animate highly deformable bodies. We gear the method towards fluid simulation by deriving the force density fields directly from the Navier-Stokes equation and by adding a term to model surface tension effects. In contrast to Eulerian grid-based approaches, the particle-based approach makes mass conservation equations and convection terms dispensable which reduces the complexity of the simulation. In addition, the particles can directly be used to render the surface of the fluid. We propose methods to track and visualize the free surface using point splatting and marching cubes-based surface reconstruction. Our animation method is fast enough to be used in interactive systems and to allow for user interaction with models consisting of up to 5000 particles.

Event-driven queue-based traffic flow microsimulation

Simulating traffic flow is an important problem in transport planning. The most popular simulation approaches for large-scale scenarios today are aggregated models. Unfortunately, these models lack temporal and spatial resolution. But microsimulations are interesting for traffic flow simulation, as they can accurately simulate features requiring both high temporal and high spatial resolution, including traffic jams and peak periods. However, most microscopic approaches involve high computational costs and expensive large computers to run them within a reasonable time. This work presents possibilities for reducing these costs by using a queue-based model and an event-driven approach jointly. The approach makes it possible to run large-scale scenarios with more than 1 million simulated person-days on networks with 10,000 links in less than 10 min on single CPU desktop computers present in most offices today.

Q-Learning for Flexible Learning of Daily Activity Plans

Q-learning is a method from artificial intelligence to solve the reinforcement learning problem (RLP), defined as follows. An agent is faced with a set of states, S. For each state s there is a set of actions, A(s), that the agent can take and that takes the agent (deterministically or stochastically) to another state. For each state the agent receives a (possibly stochastic) reward. The task is to select actions such that the reward is maximized. Activity generation is for demand generation in the context of transportation simulation. For each member of a synthetic population, a daily activity plan stating a sequence of activities (e.g., home-work-shop-home), including locations and times, needs to be found. Activities at different locations generate demand for transportation. Activity generation can be modeled as an RLP with the states given by the triple (type of activity, starting time of activity, time already spent at activity). The possible actions are either to stay at a given activity or to move ...

Performance improvements for large-scale traffic simulation in MATSim

In contrast to aggregated macroscopic models of traffic simulation, multi-agent microscopic models, such as MATSim, enable modeling of individual behavior and facilitate more detailed traffic analysis. However, such detailed modeling also leads to an increased computational burden, such that simulation performance becomes critical.

Continuous Activity Planning for a Continuous Traffic Simulation

This paper introduces a microscopic traffic simulation that continuously simulates activity7 based agent behavior and the resulting traffic. It drops iterative optimization, that builds on 8 stochastic user equilibria, and moves to a continuous planning approach. The behavioral model 9 of this approach utilizes the concept of needs to model continuous demands. Several intuitive 10 parameters control demand and facilitate calibration of versatile behaviors. These behaviors 11 originate from a planning heuristic which makes just in time decisions about upcoming activ12 ities an agent should execute. The planning heuristic bases its decisions on the current need 13 levels of an agent and the development of these levels in the near future. We illustrate the model 14 through simulation runs and suggest directions of future research. 15 Märki, F., Charypar, D. and Axhausen, K.W. 2 INTRODUCTION Microscopic travel demand simulation software (e.g. (1)) simulates people individually. Each 16 virtual person, referred to as agent, specifies day-plans which consist of activities. These activ17 ities are then executed with a simulation software. Such simulation runs provide agents with 18 feedbacks about the utility of their day-plans. The aim of each agent is to maximize its utility 19 and as a consequence, they adapt their day-plans according to the previous simulation results. 20 This replanning step is repeated until the simulation reaches a stochastic user equilibrium where 21 travel demand and travel supply are consistent (2). 22 The design of the above-mentioned approach leads to high computational complexity which 23 limits the simulation horizon of standard size scenarios to a single day. This makes it difficult to 24 investigate effects that occur between days or between weeks. Another limitation is that agents 25 have to commit themselves to a specific day-plan. This restricts the ability of agents to react 26 to unexpected events. Even worse, the repetitive nature of the simulation provides agents with 27 the knowledge of an unexpected event. Accordingly, agents adapt their day-plans in foresight 28 of that event which is impossible in real life. These shortcomings keeps us from utilizing this 29 approach for our investigations. As a consequence, a different simulation becomes necessary. 30 We propose a microscopic travel demand simulation which is capable of modeling demand 31 continuously. We use a behavioral model which utilizes the concept of needs to model the 32 continuous demand. Agents continuously track their need levels and provide this knowledge to 33 a planning heuristic which makes decisions about upcoming activities agents should execute. 34 This makes it possible, that agents can spontaneously react to unexpected events. At the same 35 time, it also reduces memory consumption because agents do not need to keep track of day36 plans. We present different calibration mechanisms which extend the need-based approach to 37 enable distinct behavior based on the day of the week or other constraints like public holidays. 38 These extensions make the need-based approach applicable to model not only recurrent activi39 ties but also the richer contents of everyday life. We plan to take advantage of developments in 40 distributed computation to decrease computation time further. The proposed activity planning 41 module will be embedded in a distributed computation environment (3). This helps to over42 come the burden of long computation times but also introduces constraints to the model. We 43 illustrate how we respect these constraints and appropriately design our algorithms. 44 The remainder of this paper is structured as follows: First, we discuss the dynamics of the 45 need-based behavior. We then introduce the structure of our need-based model and its calibra46 tion mechanisms. The next section describes the planning heuristic and its key features. This 47 is followed by an explanation of the model calibration and its validation on several examples. 48 We conclude the paper with an outlook on coming tasks. 49 OTHER WORK Arentze and Timmermans introduce need-based theory (4) and propose a model for activ50 ity generation (5) which assumes that utilities of activities are a dynamic function of needs. 51 Whereas Arentze and Timmermans use needs to generate day-plans for a multi-day planning 52 period, we use needs to make just in time decisions about upcoming activities agents should 53 execute. Kuhnimhof and Gringmuth (6) use a different approach and simulate a 7-day model 54 by recycling existing schedules. They could show that their approach produces realistic day 55 plans but suffers from a certain inflexibility when agents should spontaneously react to unex56 pected events. Charypar and Nagel (7) formulate the planning procedure as a reinforcement 57 learning problem and report that this approach has a poor performance for large scenarios. We 58 Märki, F., Charypar, D. and Axhausen, K.W. 3 try to overcome this problem by using a planning heuristic which approximates the optimal so59 lution. Schlich (8) considers travel as a consequence of people’s endeavor of need satisfaction 60 and Schönfelder (9) sees the satisfaction of needs as an explanation of the rhythms of individ61 ual travel behavior. Charypar et al. (10) address the shortcomings of the existing equilibrium 62 model and propose a need-based activity planning system. 63 NEED DRIVEN BEHAVIOR This section introduces the need driven behavior. The first subsection is a lightweight overview 64 and meant for readers who want to get a general understanding of the topic. The second sub65 section introduces the mathematics use to describe the development of need levels over time. 66