User Feedback in Latent Space Robotic Skill Learning

Pahič, Rok; Lončarević, Zvezdan; Ude, Aleš; Nemec, Bojan; Gams, Andrej

doi:10.1109/humanoids.2018.8624972

Cited by 7 publications

(5 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The right top plot in Figure 7 verifies the benefits of using AE latent space over full CDMP space by showing the worst case of 25 learning iterations. The results presented in Figures 6 and 7 are consistent with the literature where it has been reported that learning is faster in a low-dimensional latent space [5]. However, more experiments are needed to confirm that learning in AE latent spaces is faster also in general.…”

Section: Rl In Ae Latent Spacessupporting

confidence: 90%

“…Efficient data representation with autoencoders, as has already been discussed in the literature [5,17,30,31], can reduce the dimension of the learning space. In our experiments, we have shown the effectiveness of dimensionality reduction for RL of motor skills such as throwing.…”

Section: Discussionmentioning

confidence: 99%

“…It was shown in the literature that learning in autoencoder latent space outperforms learning in the full policy space [5] and learning in PCA latent space [17,18]. This paper explores the usage of statistical generalization to create the database, i.e., training samples required to train autoencoder neural networks for dimensionality reduction, thus reducing the required real-world policy executions.…”

Section: Problem Statementmentioning

confidence: 99%

“…We use this larger set to train autoencoder, and then implement learning methodologies in its latent space. Action representations and Reinforcement Learning (RL) are discussed in, for example, [5,18,30,31]. Furthermore, action generalization in feature space of a quadruped robot is discussed in [32].…”

Section: Related Workmentioning

confidence: 99%

“…This is also the case for monolithic problems where only one type of solution is possible (in our practical example, only one way of throwing) [4]. Intuitively, by reducing the dimension of the search space for learning, more successful autonomous learning algorithms can be implemented [5].…”

Section: Introductionmentioning

confidence: 98%

See 4 more Smart Citations

Generalization-Based Acquisition of Training Data for Motor Primitive Learning by Neural Networks

et al. 2021

Self Cite

View full text Add to dashboard Cite

Autonomous robot learning in unstructured environments often faces the problem that the dimensionality of the search space is too large for practical applications. Dimensionality reduction techniques have been developed to address this problem and describe motor skills in low-dimensional latent spaces. Most of these techniques require the availability of a sufficiently large database of example task executions to compute the latent space. However, the generation of many example task executions on a real robot is tedious, and prone to errors and equipment failures. The main result of this paper is a new approach for efficient database gathering by performing a small number of task executions with a real robot and applying statistical generalization, e.g., Gaussian process regression, to generate more data. We have shown in our experiments that the data generated this way can be used for dimensionality reduction with autoencoder neural networks. The resulting latent spaces can be exploited to implement robot learning more efficiently. The proposed approach has been evaluated on the problem of robotic throwing at a target. Simulation and real-world results with a humanoid robot TALOS are provided. They confirm the effectiveness of generalization-based database acquisition and the efficiency of learning in a low-dimensional latent space.

show abstract

Section: Rl In Ae Latent Spacessupporting

confidence: 90%