Useful Links
- Matt's list of EvoDesignLinks
- Reynolds has a list of evolutionary techniques applied to motion
- Reynolds has another list of individual-based models
Evolutionary Algorithms
- Genetic Algorithm
- Human-Based Genetic Algorithm - human-in-the-loop, basically what we've been using already but with the addition of human innovation
- Harmony Search Algorithm - similar to GA, but with a memory of solutions and the notion of "ensemble" to account for relationships between variables (here's a paper on this by Geem)
- Memetic Algorithm - a combination of evolutionary algorithm for global search plus individual learning at each generation for local search
- Survey of interactive evolutionary computation by Takagi (2001)
- Pareto Frontier - something to think about...
Character Animation
- Interactive Evolutionary Approach to Character Animation: A paper shockingly similar to our IED for Motion Variants paper, that I just now discovered. Differences: the genome consists of data points (at every frame) on the anim curves, splines are not included so resulting animation may lose finesse; There is no meta-designer in this system, instead objective functions and a neural net constrain the motion variation based on examples it has seen; Their system isn't integrated into a character animation software package.
- Overviews:
- Crowd Simulation - Musse and Thalmann - a book on the topic
- Groups and Crowd Simulation - Musse - a summary of crowd simulation as applied in different fields
- Steering:
- Steering Behaviors for Autonomous Characters - Reynolds - a taxonomy of behaviors for crowd motion
- A Synthetic-Vision Based Steering Approach for Crowd Simulation - Ondrej, et al (SIGGRAPH 10) - uses agent POV vision to control collision avoidance behavior using bearing angle and time-to-interaction. Impressive results... we may want to use this if it isn't too computationally expensive.
- Using Evolutionary Algorithms
- Evolving Crowd Motion Simulation with Genetic Algorithm - Wang, Li (ICARA) - uses a static fitness function instead of allowing a designer to drive the system
- Group Motion Editing:
- Group Motion Editing - Kwon, et al (SIGGRAPH 08) - a technique for editing the motion of groups using geometric constraints on a graph that represents the connectivity of characters
- Synchronized Mulit-Character Motion Editing - Kim, et al. (SIGGRAPH 09) - constraint-based optimization scheme for manipulation motions of individuals in groups quickly and precisely. This is a direct authoring approach that requires you know precisely what you want for each crowd member.
- Directable Crowd Simulation:
- Crowd Brush - Ulicny, Ciechomski, Thalmann (SCA) - They tackle most of the crowd authoring problems with brush interfaces , but they don't really address the problem of authoring behaviors . They place behavior tags on individuals and put collision objects & paths into the scene using brushes.
- Constrained Animation of Flocks - Anderson, McDaniel, Chenney (SCA) - uses forward sim to satisfy spatio-temporal constraints initially, then backward sim to meet end constraints , then uses Monte Carlo sampling to explore the behavior possibilities that meet those constraints. Meets generalized spatio-temporal constraints , but requires heavy computation (definitely off-line authoring ) and assumes that constraints and plausibility will produce "good" behaviors.
- Defeating the Authoring Bottleneck (DI Guy) - Blank, Broadbent, Crane, Pasternak (IMAGE) - programming needed to design any higher-level behaviors; finer grain control is limited to changing specific parameters for individuals (or groups) one at a time. More of a game scenario authoring tool where the author would model complex interactions explicitly (highly-intuitive scripting).
- Flow Tiles - Chenney (SCA) - Uses tile placement to design flow fields which drive agents as particles (their crowd example has a fixed flow field). Lots of limitations in actual application.
- Real-Time Crowd Simulation: A Review - Legget - a short survey of real-time crowd simulation with particular attention paid to pedestrians and particle-based methods.
- Scalable Behaviors for Crowd Simulation - Sung, Gleicher, Chenney (Eurographics) - a scalable crowd sim method that allows for situation based control (behaviors are specified spatially)
- Simulating Crowd Motion with Shape Preference and Fuzzy Rules - Chang, Li (ISALR) - crowd simulation system for path finding while maintaining a specified crowd shape
