Interactive Control of Diverse Complex Characters with Neural Networks

By Igor Mordatch, Kendall Lowrey, Galen Andrew, Zoran Popovic, E. Todorov. 2015

108 citations recorded by the source corpus.

How researchers see this paper

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

Citations in the corpus

Adaptive Online Planning for Continual Lifelong Learning (2019, background): “[19, 22] propose to use priors that make the planner stay close to policy outputs, which is problematic in changing worlds, when the policy is not accurate.”
Towards Generalization and Simplicity in Continuous Control (2017, background): “Even though the interactive tasks that we study are not as complex as those in Mordatch et al. (2015), to our knowledge, this is the ﬁrst work that considers model-free approaches for continuous control tasks where a user can interact with the learned policy.”
Collective robot reinforcement learning with distributed asynchronous guided policy search (2016, background): “Distributed systems have long been an important subject in deep learning [15], and recent work has also proposed distributed algorithms for deep reinforcement learning in the context of Q-learning and actor-critic methods [16] and learning control for simulated characters [17].”
EPOpt: Learning Robust Neural Network Policies Using Model Ensembles (2016, methodology): “Mordatch et al. (2015a) use a trajectory optimization approach and an ensemble with small finite set of models; whereas we follow a sampling based direct policy search approach over a continuous distribution of uncertain parameters, and also show domain adaptation.”
SE3-nets: Learning rigid body motion using deep neural networks (2016, background): “tensive work utilizing the MuJoCo physics engine [31] has shown how strong physics models can enable solutions to optimal control problems and policy learning in complex and contact-rich environments [11, 22]. A fundamental problem of such models is that they rely on very accurate estimates of the s”
Neural probabilistic motor primitives for humanoid control (2018, background): “Work by Mordatch et al. (2015) has shown that linear-feedback policies resulting from trajectory optimization can be used to train neural networks.”
Variance Reduction for Policy Gradient with Action-Dependent Factorized Baselines (2018, background): “Methods such as Guided Policy Search (Levine et al., 2016; Mordatch et al., 2015) and variants train policies that act on high dimensional observations like images, but use a low dimensional encoding of the problem like joint positions during the training process.”
DeepLoco (2017, methodology): “…approach uses trajectory optimization methods to provide supervision for a recurrent neural network to generate stepping movements for 3D bipeds and quadrupeds [Mordatch et al. 2015], and with the final motion coming from an optimization step rather than directly from a forward dynamics simulation.”
Robust Adversarial Reinforcement Learning (2017, background): “High-capacity function approximators such as deep neural networks have led to increased success in the ﬁeld of reinforcement learning (Mnih et al., 2015; Silver et al., 2016; Gu et al., 2016; Lillicrap et al., 2015; Mordatch et al., 2015).”
An Introduction to Deep Reinforcement Learning (2018, unknown): “demonstrated many capabilities, for instance in the case of simulated 3D bipeds and quadrupeds (e.g., Mordatch et al., 2015).”
Reinforcement learning for non-prehensile manipulation: Transfer from simulation to physical system (2018, methodology): “This is similar to a procedure outlined in [43] for training interactive and robust policies.”
Guided Policy Search as Approximate Mirror Descent (2016, methodology): “A common and simple choice for the local policies p i ( u t | x t ) is to use time-varying linear-Gaussian controllers of the form p i ( u t | x t ) = N ( K t x t + k t , C t ) , though other options are also possible [12, 11, 19].”
Learning symmetric and low-energy locomotion (2018, unknown): “n date: August 2018. Learning Symmetric and Low-Energy Locomotion • 144:3 a trajectory optimization algorithm and demonstrated interactive control of character locomotion with different morphologies [Mordatch et al. 2015]. In this work, we aim to develop an algorithm that can synthesize plausible lo”
Learning human behaviors from motion capture by adversarial imitation (2017, methodology): “Furthermore, trajectories optimized using model-based approaches can be distilled into a neural network policy [19].”
Combining model-based policy search with online model learning for control of physical humanoids (2016, methodology): “Following [16], we replace the hard equality constraint with a quadratic penalty with weight α:”
Learning locomotion skills using DeepRL: does the choice of action space matter? (2016, methodology): “2016]; velocities under an implicit control policy [Mordatch et al. 2015]; or provide abstract actions [Hausknecht and Stone 2015].”