Learning physics-based motion style with nonlinear inverse optimization

By C. K. Liu, Aaron Hertzmann, Zoran Popovic. ACM Transactions on Graphics, 2005

398 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Control and Systems Engineering

View Paper: https://www.semanticscholar.org/paper/6a0c7d16a27f673e807b79483637d89b7b539cf5

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Control and Systems Engineering

Citations in the corpus

HuMoR: 3D Human Motion Model for Robust Pose Estimation (2021, background): “Many motion models have been learned for computer animation [13, 50, 82, 52, 56, 38, 89] including recent recurrent and autoregressive models [33, 29, 36, 105, 54].”
Terrain-adaptive bipedal locomotion control (2010, methodology): “Offline methods can be used to synthesize animation of bipedal locomotion [Rose et al. 1996; Safonova et al. 2004; Liu et al. 2005].”
Feature Correspondence Via Graph Matching: Models and Global Optimization (2008, methodology): “We use the Nonlinear Inverse Optimization (NIO) algorithm described in [26].”
Optimizing Locomotion Controllers Using Biologically-Based Actuators and Objectives (2012, background): “More nuanced objectives can be learned from inverse optimization [Liu et al. 2005], but are dependent on training data.”
Gaussian Processes for Signal Strength-Based Location Estimation (2006, methodology): “GP regression has been used with great success in a variety of applications, including sensor networks [3], data visualization [9], and computer animation [12].”
Learning symmetric and low-energy locomotion (2018, methodology): “An alternative approach to obtain natural character motions is to incorporate motion data such as videos [Wampler et al. 2014] or motion capture data [da Silva et al. 2008; Lee et al. 2010, 2014; Liu et al. 2005; Muico et al. 2009; Sok et al. 2007; Ye and Liu 2010].”
Robust physics-based locomotion using low-dimensional planning (2010, methodology): “Since trajectory optimization [Witkin and Kass 1988] for full-body locomotion remains very computationally expensive [Liu et al. 2005; Wampler and Popović 2009], we optimize the trajectory of a low-dimensional preview model instead.”
QuaterNet: A Quaternion-based Recurrent Model for Human Motion (2018, background): “Alternatively to the direct modeling of joint rotations/positions, physics-inspired models of the human body have also been explored [33] but such models have been less popular for generation with the availability of larger motion capture datasets [1, 39].”
Estimating contact dynamics (2009, background): “Physics-based animation with spring-based contact models is common in computer graphics [11, 14, 20].”
From human to humanoid locomotion—an inverse optimal control approach (2010, background): “Liu et al. (2005) study realistic generation of character motion by physics-based models.”
Inverse Dynamics for Action Recognition (2013, methodology): “We used a similar biomechanical model as used in [23].”
Two Distributed-State Models For Generating High-Dimensional Time Series (2011, methodology): “Models based on the physics of masses and springs have produced someimpressive results by using sophisticated “energy-based” learning methods (LeCun et al., 1998) to estimate physical parameters from motion capture data (Liu et al., 2005).”
Optimal gait and form for animal locomotion (2009, methodology): “A particularly interesting possibility for determining these tradeoffs would be to employ inverse optimization techniques such as those used to find the values for passive elements by [Liu et al. 2005].”
A Physics-Based Model Prior for Object-Oriented MDPs (2014, methodology): “, 2002), and even humanoid motion (Liu et al., 2005).”
Category Modeling from Just a Single Labeling: Use Depth Information to Guide the Learning of 2D Models (2013, methodology): “Supervised uses the ground truth, instead of 3D matching assignments, to guide the model learning, whereas Supervised+NIO uses nonlinear inverse optimization (NIO) for model learning [4, 38].”