Discovery of complex behaviors through contact-invariant optimization

By Igor Mordatch, E. Todorov, Zoran Popovic. 2012

558 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Control and Systems Engineering, Reinforcement Learning

How researchers see this paper

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

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Control and Systems Engineering

In Reinforcement Learning

Citations in the corpus

Emergence of Locomotion Behaviours in Rich Environments (2017, background): “For instance, [24] show sophisticated skill sequencing for maneuvering obstacles on a parametric terrain, while [25, 26, 27] demonstrate how terrain adaptive behaviors or other skilled movements can emerge as the result of optimization problems.”
Towards Generalization and Simplicity in Continuous Control (2017, background): “The complexity of the systems being controlled is not yet at the level of what can be achieved through trajectory optimization in simulation (Tassa et al., 2012; Mordatch et al., 2012; Al Borno et al., 2013) or with hand-crafted controllers on physical robots (e.g. Boston Dynamics).”
Differentiable Physics and Stable Modes for Tool-Use and Manipulation Planning (2018, methodology): “Note that in CIO [21], all contacts were in fact assumed to be stable modes.”
Toward an Integration of Deep Learning and Neuroscience (2016, background): “Computationally, we now know that optimization of trajectories gives rise to elegant solutions for very complex motor tasks (Harris and Wolpert, 1998; Todorov and Jordan, 2002; Mordatch et al., 2012).”
Motion planning with sequential convex optimization and convex collision checking (2014, background): “Finally, there recently has been considerable progress in trajectory optimization for dynamical systems [32, 26, 39, 51, 12].”
Grasp'D: Differentiable Contact-rich Grasp Synthesis for Multi-fingered Hands (2022, methodology): “Inspired by Contact-Invariant Optimization [64, 65] we relax the problem making it more forgiving to gradient-based optimization.”
Catch & Carry (2019, unknown): “For many physics-based graphics approaches involving object interactions, a shortcut is often taken by forming fixed attachments between the hands of the body and the manipulated object in order to simplify the problem [Coros et al. 2010; Mordatch et al. 2012; Liu et al. 2012; Peng et al. 2019].”
Robust Recovery Controller for a Quadrupedal Robot using Deep Reinforcement Learning (2019, methodology): “Optimization-based methods [17] have a hard time to solve”
Modeling, learning, perception, and control methods for deformable object manipulation (2021, methodology): “MuJoCo is a proprietary engine that has been used for modeling rigid-linked characters ( 30 ) or as the environment backend of re-inforcement learning (RL) benchmarks ( 31 ).”
Finding Locally Optimal, Collision-Free Trajectories with Sequential Convex Optimization (2013, background): “Finally, there recently has been considerable progress in trajectory optimization in situations with contact, some of the most notable results include the ones described by Mordatch et al. [20], Posa et al. [23], and Tassa et al. [28].”
Combining model-based policy search with online model learning for control of physical humanoids (2016, methodology): “The existing approach we use is Contact-Invariant Optimization (CIO) [18], which is a direct trajectory optimization method based on an inverse dynamics model.”
Neural state machine for character-scene interactions (2019, background): “2012; Lo and Zwicker 2008] and Maximum A Posteriori (MAP) estimation [Min and Chai 2012], local optimization [Mordatch et al. 2012] and stochastic optimization.”
DeepMimic (2018, methodology): “Trajectory optimization has been explored for synthesizing physically-plausible motions for a variety of tasks and characters [Mordatch et al. 2012; Wampler et al. 2014].”
An Introduction to Trajectory Optimization: How to Do Your Own Direct Collocation (2017, methodology): “These methods are covered in detail in [47], [46], and [39].”
Interactive Control of Diverse Complex Characters with Neural Networks (2015, methodology): “While these feedback gains are a byproduct of indirect trajectory optimization methods such as LQG, they are not an obvious result of direct trajectory optimization methods like CIO.”
A direct method for trajectory optimization of rigid bodies through contact (2014, background): “In another recent approach, Mordatch et al.[13] have also explored relaxations of the the complementarity contact constraints in the context of trajectory optimization.”
Learning symmetric and low-energy locomotion (2018, methodology): “Apart from designing finite state machines or using motion data, reward engineering combined with trajectory optimization has also been frequently applied to generate physically-based character motions [Al Borno et al. 2013; Ha and Liu 2014; Mordatch et al. 2012, 2013; Wampler and Popović 2009].”