Report

A predictive Collision Avoidance Model for Pedestrian Simulation Author: Ioannis Karamouzas et al. Presented by: Jessica Siewert Content of presentation • • • • • • Previous work The method Implementation Experiments Assessment Developments since Introduction – Previous work • • • • Dynamic potential-field approach (too general) Corridor-Map-Method Helbing Social Force Fields Example-based (too expensive) Introduction – Now we want… • Anticipation and prediction (so in advance) • Deal with large and cluttered environments • No constant change of orientation, pushing each other and moving back/forth Introduction – We got… • Reynolds unaligned collision avoidance • => Feurtey predicts potential collisions within time and resolves by adapting speed and trajectory • => Paris et al. Anticipative model to steer • Shao and Terzopoulos: Reactive routines to determine avoidance maneuvers. Introduction – We got… • Van den Berg Reciprocal Velocity Obstacle • Pettré et al. Egocentric model for local collision avoidance Introduction – Our method… • Based on force field approach • Early avoidance hypothesis, anticipation/prediction • Energy-efficient motions – Less curved paths – Smooth natural flow – Oscillation-free Introduction – Contributions… • • • • Force field method based (Shao, Berg, Pettré don’t) Easier in formulation and implementation Faster, able to handle thousands Calculated differently producing better looking results (visually pleasing, smoothly avoiding) The method – Overview • Pedestrian Interactions • => Pedestrian Simulation Model • Collision Avoidance The method – Pedestrian Interactions • Scanning and Externalization • Personal Space • Principle of Least Effort The method – Pedestrian Sim. Model • Modeled as little cylinders with radius r • The pedestrian tries to reach its goal • The goal is pulling the pedestrian towards itself with a goal force The method – Pedestrian Sim. Model • The pedestrian wants to move at a certain speed • It reaches this spreed gradually over time The method – Pedestrian Sim. Model • All the walls act on the pedestrian repulsively • D shortest distance between P and wall • D safe discance P likes from the wall iw s The method – Pedestrian Sim. Model • A pedestrian keeps a distance from others to feel comfortable (“Personal space”) • Modeled as a disc with radius p>r (is varied) http://www.mysocalledsensorylife.com/?p=2021 The method – Pedestrian Sim. Model • The collision occurs when another pedestrian Pj comes in the personal space of Pi at time tc The method – Pedestrian Sim. Model • A pedestrian has an anticipation time (can vary) • Collisions within this time are actively avoided • To simulate this an evasive Force is applied Collision avoidance • Collision prediction Collision avoidance • Selecting pedestrians – Sorted on increasing collision time – Keep the first 2 to 5 Collision avoidance • Avoidance maneuvers Collision avoidance • Computing the evasive Force – Weighted sum of N forces – OR – Iterative approach! Agile101.net Implementation • Efficient Collision Prediction – Anticipation time – Iterative approach – Vary p, r, v and t – Maximum distance Implementation • Adding variation – Noise Force • Time integration – Simulation time steps – Sum of forces – Orientation Experiments – Claim recall • • • • Anticipation/prediction based Easier in formulation and implementation Faster, able to handle thousands Energy-efficient motions – Less curved paths – Smooth natural flow – Oscillation-free – Visually pleasing/natural looking Movies… • • • • • • • file:///C:/Users/Jessica/Downloads/Circle.avi file:///C:/Users/Jessica/Downloads/Scene0.avi file:///C:/Users/Jessica/Downloads/Scene1.avi file:///C:/Users/Jessica/Downloads/Scene2.avi file:///C:/Users/Jessica/Downloads/Scene3.avi file:///C:/Users/Jessica/Downloads/park.avi file:///C:/Users/Jessica/Downloads/crosswalks .avi Assessment – promises • Scanning and externalization? • Natural looking? • Easy implementation: extendability? Assessment – method • Reasoning that leads to smart pedestrian selection • Reasoning that leads to iterative approach • How would this method combine with obstacle avoidance methods? Assessment – experiments • 25% of CPU usage? • What about the high-cluttered environments? • How is the time step chosen? Assessment – results • Swirl effect • Up front anticipation results in no interaction • No ellipse-shaped personal space needed? Assessment – shortcomings • No couples or coherent groups • No cultural, cognitive or psychological factors • Nothing like the reciprocal method Developments since then • Path Planning for Groups Using Column Generation (Marjan van den Akker, Roland Geraerts e.a.) • http://gamma.cs.unc.edu/PLE/pubs/PLE.pdf • http://d.wanfangdata.com.cn/periodical_zggd xxxswz-jsjkx201003011.aspx • http://people.cs.uu.nl/ioannis/publications.ht ml