Bayesian optimization for learning gaits under uncertainty

By Roberto Calandra, André Seyfarth, Jan Peters, Marc Peter Deisenroth. 2015

352 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Biomedical Engineering, Computational Theory and Mathematics

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 Biomedical Engineering

In Computational Theory and Mathematics

Citations in the corpus

Sim-to-Real: Learning Agile Locomotion For Quadruped Robots (2018, background): “Bayesian optimization is often data efficient enough to be applied directly on real robots [14, 15].”
Maximizing acquisition functions for Bayesian optimization (2018, background): “2: for t = 1 ,...,T do : Givenmodel ,acquisition , : Givenmodel M ,acquisition L”
Deep Kernels for Optimizing Locomotion Controllers (2017, methodology): “[1] use a smaller biped with a ﬁnite-state-machine controller, which is not widely used.”
RMA: Rapid Motor Adaptation for Legged Robots (2021, background): “The input channel number, output channel number, kernel size, and stride of each layer are [32, 32, 8, 4], [32, 32, 5, 1], [32, 32, 5, 1].”
Learning to Walk via Deep Reinforcement Learning (2018, background): “Real-world applications typically make use of simple and inherently stable robots [14] or low-dimensional gait parameterizations [8, 29, 36], or both [45].”
Reset-free Trial-and-Error Learning for Robot Damage Recovery (2016, methodology): “A popular alternative is to view RL as an optimization of the parameters of a policy [16], which can be solved with gradient-based methods [26, 10], evolutionary algorithms [27] or other optimization methods like Bayesian optimization [7, 28, 29].”
CLOCs: Camera-LiDAR Object Candidates Fusion for 3D Object Detection (2020, background): “Compared to 2D object detection, which has been well-studied [1], [2], [3], [4], 3D object detection is more challenging with more output parameters needed to specify 3D oriented bounding boxes around targets.”
Scalable Global Optimization via Local Bayesian Optimization (2019, background): “The global optimization of high-dimensional black-box functions—where closed form expressions and derivatives are unavailable—is a ubiquitous task arising in hyperparameter tuning [42]; in reinforcement learning, when searching for an optimal parametrized policy [7]; in simulation, when calibrating ”
Model Identification via Physics Engines for Improved Policy Search (2017, background): “Policy search can also be achieved through Bayesian optimization and has been used for gait optimization [14], where Central Pattern Generators are a popular policy parameterization [15].”
Bayesian Optimization for Adaptive Experimental Design: A Review (2020, background): “Recent years have seen new applications in areas such as robotics [45], [46],”
BoTorch: Programmable Bayesian Optimization in PyTorch (2019, background): “, 2019), robotic control (Calandra et al., 2016; Antonova et al., 2017), chemical design (Griffiths & Hernández-Lobato, 2017; Li et al.”
Bayesian Optimization is Superior to Random Search for Machine Learning Hyperparameter Tuning: Analysis of the Black-Box Optimization Challenge 2020 (2021, background): “For example, chemical engineering, materials discovery, manufacturing design, control systems and drug discovery are also effective uses of BO [8, 9, 20, 23, 26, 33, 36, 49, 64].”
Learning manipulation skills from a single demonstration (2018, methodology): “For example, Bayesian optimization was used to learn gait parameters for locomotion skills (Lizotte et al. 2007; Calandra et al. 2015).”
Deep reinforcement learning for tensegrity robot locomotion (2016, methodology): “Robotic learning methods have previously produced successful policies for tasks such as locomotion for bipeds and quadrupeds [2], [3], [4], [5], [6], [7].”
Black-box data-efficient policy search for robotics (2017, background): “, policy gradient algorithms [7], [8] or Bayesian optimization [2], [9], [10]) which only use the (cumulative) reward at the end of each episode.”
Data-efficient neuroevolution with kernel-based surrogate models (2018, background): “Modern surrogate-assisted optimization often takes place within the framework of Bayesian optimization (BO) [2, 4, 8, 11, 17, 25].”
Policies Modulating Trajectory Generators (2018, methodology): “Popular techniques include Black-box Optimization [1], Bayesian Optimization [2], Evolutionary Algorithms [3, 4] and Reinforcement Learning [5, 6, 7].”