Articulated Swimming Creatures

Contribution:
Evaluation:
Reproducibility:
Improvements:

Contribution: The paper presents a simulation environment for articulated virtual creatures swimming in a fluid environment, and an optimization method to find viable swimming "gaits" for any given creature morphology. The simulator uses a voxelization approach along with standard Navier-Stokes equations to simulate the interaction between a creature and its fluid environment. Covariance Matrix Adaptation is then used to optimize control of a given creature as it swims. The results tend to resemble real-world creature movement due to the accurate simulation and effective objective function used in optimization.

Evaluation: The system is evaluated by testing it on models of several different real-world swimming creatures, including a fish, eel, frog and manta ray. Success is measured as how well the resulting motion approximates real motion observed in nature, in addition to how well the creatures can follow a path. Finally the same test is run on a creature morphology that does not exist in nature, with subjective observations on plausibility of the motion made instead of comparison to (non-existent) real-world motion. They also ran the same set of experiments using a much simpler fluid solver and compared the results.

Reproducible: The paper is reproducible due to the high level of detail presented on both how their simulator works and how their optimization routine functions. Almost all of the equations used in simulation are presented. The optimization is not presented in quite as much detail, but the gaps should be fairly easy to fill in by reading about the Covariance Matrix Adaptation method.

Improvement: I have no complaints about this paper. It is very well-organized, well-written and extremely thorough, making it relatively easy to understand the concepts presented within. It is also very thorough in its evaluation and testing of the system.

-- Main.cdoran - 01 Dec 2011

Contribution: The authors attempt to animate swimming creatures while seeking some measure of physical accuracy. To do so, they create a variety of virtual articulated creatures (both based on real animals and purely invented) and optimize control strategies while also simulating fluid dynamics.

Evaluation: Their ambitious goal is hard to evaluate from the onset, and sure enough their results are of "oh, look how pretty it looks" variety. Some of the remarks do make sense, like the fact that the opmitized control strategies ressemble those of real creatures, so in that sense it seems obvious that their approach is reasonable.

Reproducibility: As usual with these papers, the equations seem to be worked out in sufficient details. Not too much detail is given on the experiments, and in particular not too much is mentioned on how the optimization was carried out, which means it could have been either extremely easy or extremely hard.

Improvements: They mention themselves that the voxelization approach is not ideal, and can be improved upon. But all in all, they seem to have succeeded in their goal (which is why their suggested future work doesn't really make sense, swimming at the surface of the water seems like quite a different problem).

-- Main.ginestra - 01 Dec 2011

Contribution:
There are two main contributions of the study. First, they propose an optimization solution for the control of swimming motions of several creatures. Second, their method considers coupling between creature motions and the fluid interactions.

Evaluation:
The study presents results of swimming motions for a variety of animals that have different configurations and constraints. They also successfully tested their method for controlling an artificial creature. They suggest that the resulting motions look natural, and for the short-bodied fish, and the eelm, they are in agreement with the laboratory data in terms of the generated vortex trails. For other creatures, they cannot make a comparison as there is not any data for them.

Reproducibility:
Most of the equations are given and explained throughout the paper. Some of the parameters are also given, e.g. weight values, etc. However there are some parameters that they manually set, e.g. energy bound, which are not stated in the paper. Hence, it's pretty much reproducible but requires some hand tuning of some of the parameters.

Improvements:
The study can be further extended by investigating different functions for the optimization, other than just using a sine function. By that way, they can try to optimize more sophisticated motions, which can then be generalized to generate combinations of those motions. Paper is well-structured, and easy to follow.

-- Main.ooguz - 01 Dec 2011

---

Contribution: This paper presents a complete system for controlling a wide variety of aquatic animals in a simulated fluid environment by creature/fluid simulation and optimization of the creature motion parameters. It is able to automatically find the optimal locomotion for an aquatic animal represented by an articulated rigid body system without prior knowledge the animal’s behavior.

Evaluation: The study is evaluated by demonstrating optimized swimming gaits of different aquatic animals and swimming strategies. They also compare the motion in a Navier-Stokes fluid with the motion in a simplified fluid.

Reproducibility: Given the sufficient details presented by the paper, it shouldn’t be hard to reproduce it especially when most of the equations are explained. I found the path following may be a little hard to implement.

Improvement: Overall the paper is very well structured and easy to follow. I can read through without going back and forth. In my opinion the paper does a good job in clearing up confusion. As an example, after watching the video I had questions about why the creatures seem to use so much energy to just make a little move. And I found this is explained well in the ‘limitation’ section.

Baoxuan - 01 Dec 2011

---

** Contribution? Provides a tool that can produce physcially realistic aquatic motion for a wide array of creatures. So they don't have to be animated manually.

** How are the results evaluated? They build different creature models and see how they swim with or without defined paths.

** Reproducible? Highly likely. The equations for simulations are given. Controller equations are given, but not the gains and dampling coefficients. Equations to simulate two-way coupling are given. Weights of the objective function terms are given.

** Improvements? The swimming motions don't look very realistic to me. And they probably should compare the motion with real footages in order to make the paper more convincing.

-- Main.shuoshen - 01 Dec 2011

Contibution: The paper presents an approach creating realistic swimming behaviour given the model of a body. The idea of the paper is simple: the liquid is simulated by the Navier-Stokes/Euler equations, and boundary conditions are given by the swimming creature. The gait of the creature is optimized in terms of efficiency (speed) and staying within certain limits of energy consumption.

Evaluation: It's really hard to evaluate appropriately, since it's hard (if possible at all?) to do mocap on real swimming creatures. So the evaluation is confined to "we tried this model and it looks good" - which is acceptable in this case.

Reproducibility: yes, possible to reproduce. It doesn't look like there anything is the paper that won't work, everything is very straightforward.

Improvements: Even though they did state their main limitation in the paper - the parameterization of the joint space, they didn't really elaborate on how important that is. As I understand, if we presume that the gait is parameterized by one sine function, the motion is 'static', i.e. the optimization cannot take into account any obstacles, other streams in the fluid, etc. - basically, everything that may make the motion uneven.

-- MikhailBessmeltsev - 01 Dec 2011

---

Contribution: This paper presents a method for optizing the swimming performance of a variety of creatures in a virtual aquatic environment that simulates the two-way force couping between the fluid and rigid segments of the creatures. Their objective function They utilize Covariance Matrix Adaptation (CMA) for their optimization stage, in addition to two heuristics used to accelerate convergence to a minimum for the objective function.

Evaluation: The authors describe the time necessary to optimize the creature stroke (hours) and to simulate each frame of the fluid environment (seconds) for a variety of creature morphologies. They present qualitative analysis of the resulting swimming styles compared to observed styles of real creatures. Additionally, they qualitatively compare the swimming styles of the creatures using either the sophisticated Navier-Stokes implementation or a simple normal-force fluid model. They particularly note that the computationally expensive NS model is required for creatures that swim by jet propulsion.

Reproducibility: The authors provide sufficient details on their PD controllers and objective functions. The coupled solid-fluid environment is able to be reproduced by referencing the original source paper for the technique. Given the relatively straightforward objective of the paper and the use of techniques described in earlier work in fluid simulation and optimization, the paper certainly provides a reproducible basis.

Improvements:

• Clarification: How do the authors choose the period and phase offset values for the reference trajectories sine functions? Does the phase offset vary per degree of freedom for creatures beside the accordion?
• Why are only 2 swim cycles used per optimization iteration? Is it possible that irregularities or problems would result for a creature after >2 cycles, especially given the dynamic forces given from the fluid simulation?

-- BenHumberston - 01 Dec 2011

Contribution: The authors develop a system that balances between physical realism and generality in control strategies of aquatic animals. Manual tuning of parameters and required a priori knowledge of the animals' behaviour is minimized, allowing the system to generalize to arbitrary creature models.

Evaluation: Again, success of control strategies developed are based on subjective aesthetics of the resulting movements. In terms of accomplishing the task of path planning through the environment, the ability for the creature to follow a given path is also evaluated and was deemed a success.

Reproducibility: Exceptionally well detailed in terms of formulas, parameters and implementation.

Improvements: It would have been nice to see a quantitative evaluation of the system (i.e. statistical comparison to movement strategies of actual aquatic creatures)

-- DanielTroniak - 01 Dec 2011

---

-- MichielVanDePanne - 27 Nov 2011

What is the contribution of the paper?

This paper found a way to automatically decide the optimal locomotion in the aquatic environment. First, in this paper, it first simulated the motion in water controlled by N-S equation. And then, for locomotion, the motion of joint with most efficiency is used. That is easy to understand, the aquatic animal always use the easiest and most efficiency way to move utilizing the help of force in water. Furthermore, with the optimization control by CMA, it simulates more nature locomotion and get a balance between physical realism and generality.

How are the results evaluated?

This paper animates a wide variety of aquatic animals, such as clownfish, turtle, frog and so on. And then evaluating motion, this paper demonstrates the effectiveness in different gaits. Furthermore, the author also pointed out that with different kinds of fluid flow, this system will produce dramatically different animating results.

Is the paper reproducible?

Partially reproducible. The dataset and the model is easy to reproduce. And the N-S simulation is such a mature technique that we can implement it easily by learning some course paper. But for the implementation of CMA, they only points out that the optimization is based on the CMA. The detailed parameters are missing. And the source code is invisible.

How could the paper research or paper writing be improved?

Considering the optimization is the second contribution in this paper, I think it should have more introductions about optimization (i.e. how to use CMA here) by details. Although the reader can learn CMA by referring the corresponding paper, the paper should also show the detailed implementation of CMA in the paper case.

-- Main.chuanzhu - 02 Dec 2011

---

Contribution:

In previous, most scientific models for swimming motion are customized to specific species with predefined locomotion patterns. This paper describes a complete system for controlling a wide variety of aquatic animals in a simulated fluid environment.

How are the results evaluated:

Use particle traces to visualize the fluid flow, and use several types of body shape and styles for fishes to "swim" in the system. It also shows the comparison when doing different CMA iterations. The result shows that users can define a supposing track for animals to chase and it shows a good result.

Reproducability:

According to their clear simulation work and optimization functions, we can reproduce the system in a sort of clear way. Still some parameters are manually set and not mentioned in paper, however, it is not so hard to explore them out.

Improvement:

One question about the viscous fluid. If viscosity is a problem, why not make viscosity as a function of penalty for optimization. Everything else are good enough I think.

--Main.Jingxian Li - 02 Dec 2011