Dynamic movement primitives in latent space of time-dependent variational autoencoders

Chen, Nutan; Karl, Maximilian; Smagt, Patrick van der

doi:10.1109/humanoids.2016.7803340

Cited by 52 publications

(67 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…While AEs and VAEs can be used to learn the whole-body human posture, they cannot be used to represent whole-body trajectories over time in a smooth and coherent manner (i.e., without jolts) since there is no postural time-dependence. This issue was well explained by Chen et al in [8], which proposed to force the temporal dependency by learning DMPs in the latent space. Their method, called VAE-DMP, uses Deep Variational Bayes Filters (DVBF) [24], where Bayesian filtering is applied on latent variables with temporal dependencies, with a recurrent deep neural network composed of chained VAEs An improvement of VAE-DMP, called Variational Time Series Feature Extractor (VTSFE), was proposed by Chaveroche et al in [5] to encode features of the time series for both classification and prediction purposes.…”

Section: B Dimensionality Reduction In a Latent Spacementioning

confidence: 94%

“…AE-ProMPs is computationally efficient and suitable for our application. We compare it with similar methods proposed to encode whole-body movements in a latent space, namely VAE-DMP [8] and VTSFE [5]. The first exploits variational autoencoders (VAE) to compress the movement in a reduced latent space, then forces the continuity of the latent space trajectories using Dynamic Motion Primitives (DMPs).…”

Section: Introductionmentioning

confidence: 99%

“…While both these methods are very interesting for their capacity to produce a coherent latent space that preserves the continuity of the trajectories, they are computationally expensive. Moreover, the complexity induced by the dynamic forcing function can be skipped in our case since we do not consider individual latent space trajectories, as in [8], but learn probability distributions over the latent space trajectories.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Prediction of Human Whole-Body Movements with AE- ProMPs

Denny

Chaveroche

Colas

et al. 2018

2018 IEEE-RAS 18th International Conference on Humanoid Robots (Humanoids)

View full text Add to dashboard Cite

The ability to predict the future intended movement is crucial for collaborative robots to anticipate the human actions and for assistive technologies to alert if a particular movement is non-ergonomic and potentially dangerous for the human health. In this paper, we address the problem of predicting the future human whole-body movements given early observations. We propose to predict the continuation of the high-dimensional trajectories mapped into a reduced latent space, using autoencoders (AE). The prediction is based on a probabilistic description of the movement primitives (ProMPs) in the latent space, which notably reduces the computational time for the prediction to occur, and hence enables to use the method in real-time applications. We evaluate our method, named AE-ProMPs, for predicting future movements belonging to a dataset of 7 different actions performed by a human, recorded by a wearable motion tracking suit.

show abstract

Section: B Dimensionality Reduction In a Latent Spacementioning

confidence: 94%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Prediction of Human Whole-Body Movements with AE- ProMPs

Denny

Chaveroche

Colas

et al. 2018

2018 IEEE-RAS 18th International Conference on Humanoid Robots (Humanoids)

View full text Add to dashboard Cite

show abstract

“…An early attempt aimed to link neural networks to path planning by specifying obstacles into topologically ordered neural maps and using neural activity gradient to trace the shortest path, with neural activity evolving towards a state corresponding to a minimum of a Lyapunov function [11]. More recently, a method was developed that enables the representation of high dimensional humanoid movements in the low-dimensional latent space of a time-dependent variational autoencoder framework [12]. Reinforcement Learning (RL) approaches have also been proposed for motion planning applications [13], [14].…”

Section: Related Workmentioning

confidence: 99%

Neural Path Planning: Fixed Time, Near-Optimal Path Generation via Oracle Imitation

Bency

Qureshi

Yip

2019

2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

View full text Add to dashboard Cite

Fast and efficient path generation is critical for robots operating in complex environments. This motion planning problem is often performed in a robot's actuation or configuration space, where popular pathfinding methods such as A*, RRT*, get exponentially more computationally expensive to execute as the dimensionality increases or the spaces become more cluttered and complex. On the other hand, if one were to save the entire set of paths connecting all pair of locations in the configuration space a priori, one would run out of memory very quickly. In this work, we introduce a novel way of producing fast and optimal motion plans for static environments by using a stepping neural network approach, called OracleNet. Ora-cleNet uses Recurrent Neural Networks to determine end-toend trajectories in an iterative manner that implicitly generates optimal motion plans with minimal loss in performance in a compact form. The algorithm is straightforward in implementation while consistently generating near-optimal paths in a single, iterative, end-to-end roll-out. In practice, OracleNet generally has fixed-time execution regardless of the configuration space complexity while outperforming popular pathfinding algorithms in complex environments and higher dimensions 1 .

show abstract

“…valid regions in the original continuous space; the states having low posterior probabilities (or equivalently, having high costs in the planning problem) up to the current time-step tend not to be re-sampled (line [12][13]. After the forward recursion, the algorithm picks the most likely final state y * K (line [15][16], and then constructs the whole trajectory with backtracking by looking at its ancestry (line [17][18][19][20].…”

Section: Dynamic Programming For Computing Map Trajectory Using Pamentioning

confidence: 99%

Approximate Inference-Based Motion Planning by Learning and Exploiting Low-Dimensional Latent Variable Models

Chae

Choi

2018

IEEE Robot. Autom. Lett.

View full text Add to dashboard Cite

This work presents an efficient framework to generate a motion plan of a robot with high degrees of freedom (e.g., a humanoid robot). High-dimensionality of the robot configuration space often leads to difficulties in utilizing the widely-used motion planning algorithms, since the volume of the decision space increases exponentially with the number of dimensions. To handle complications arising from the large decision space, and to solve a corresponding motion planning problem efficiently, two key concepts are adopted in this work: First, the Gaussian process latent variable model (GP-LVM) is utilized for low-dimensional representation of the original configuration space. Second, an approximate inference algorithm is used, exploiting through the duality between control and estimation, to explore the decision space and to compute a high-quality motion trajectory of the robot. Utilizing the GP-LVM and the duality between control and estimation, we construct a fully probabilistic generative model with which a high-dimensional motion planning problem is transformed into a tractable inference problem. Finally, we compute the motion trajectory via an approximate inference algorithm based on a variant of the particle filter.

show abstract

Dynamic movement primitives in latent space of time-dependent variational autoencoders

Cited by 52 publications

References 10 publications

Prediction of Human Whole-Body Movements with AE- ProMPs

Prediction of Human Whole-Body Movements with AE- ProMPs

Neural Path Planning: Fixed Time, Near-Optimal Path Generation via Oracle Imitation

Approximate Inference-Based Motion Planning by Learning and Exploiting Low-Dimensional Latent Variable Models

Contact Info

Product

Resources

About