Transfer learning of shared latent spaces between robots with similar kinematic structure

Delhaisse, Brian; Esteban, Domingo; Rozo, Leonel

doi:10.1109/ijcnn.2017.7966379

Cited by 26 publications

(13 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Within the BO applications that employ GPs, the prior knowledge can be transferred as GP priors (Raina et al, 2006) and hyperparameters (Perrone et al, 2019) from the trained domains to provide predictive information about the unknown features and distributions in the new test domains. In robotics, TL is typically employed as the transfer of the models of kinematics and dynamics between simulated and physical platforms of conventional rigid robotic systems, such as manipulators (Devin et al, 2017;Makondo et al, 2018), humanoids (Delhaisse et al, 2017), and quadrotor platforms (Helwa and Schoellig, 2017). However, the application of TL on soft robotics systems is still in its early infancy (Schramm et al, 2020).…”

Section: Introductionmentioning

confidence: 99%

Task space adaptation via the learning of gait controllers of magnetic soft millirobots

Demır

Çulha

Karacakol

et al. 2021

The International Journal of Robotics Research

View full text Add to dashboard Cite

Untethered small-scale soft robots have promising applications in minimally invasive surgery, targeted drug delivery, and bioengineering applications as they can directly and non-invasively access confined and hard-to-reach spaces in the human body. For such potential biomedical applications, the adaptivity of the robot control is essential to ensure the continuity of the operations, as task environment conditions show dynamic variations that can alter the robot’s motion and task performance. The applicability of the conventional modeling and control methods is further limited for soft robots at the small-scale owing to their kinematics with virtually infinite degrees of freedom, inherent stochastic variability during fabrication, and changing dynamics during real-world interactions. To address the controller adaptation challenge to dynamically changing task environments, we propose using a probabilistic learning approach for a millimeter-scale magnetic walking soft robot using Bayesian optimization (BO) and Gaussian processes (GPs). Our approach provides a data-efficient learning scheme by finding the gait controller parameters while optimizing the stride length of the walking soft millirobot using a small number of physical experiments. To demonstrate the controller adaptation, we test the walking gait of the robot in task environments with different surface adhesion and roughness, and medium viscosity, which aims to represent the possible conditions for future robotic tasks inside the human body. We further utilize the transfer of the learned GP parameters among different task spaces and robots and compare their efficacy on the improvement of data-efficient controller learning.

show abstract

Section: Introductionmentioning

confidence: 99%

Task space adaptation via the learning of gait controllers of magnetic soft millirobots

Demır

Çulha

Karacakol

et al. 2021

The International Journal of Robotics Research

View full text Add to dashboard Cite

show abstract

“…An interesting finding that requires further investigation is that, although complex non-linear mappings are generally preferred in most transfer learning problems for robotics, such as shared Autoencoders, Shared-GPLVM [16] and LPA [22,29], simple linear mappings such as Procrustes Analysis could be more beneficial in a case where human demonstrations are used to provide initialization for learning with reinforcement learning for humanoids [51,52], or a case where a human is allowed to provide feedback to the robot learner for correcting its reproductions [48][49][50]. This is because they can learn mappings from very few samples, which is desired for physical robots, and that they preserve the overall gist of the transferred skills, which would guide a reinforcement learner towards relevant spaces for exploration.…”

Section: Discussionmentioning

confidence: 99%

“…Another example is using Shared Gaussian Process Latent Variable Models (Shared-GPLVM) to jointly learn a latent representation of skills in a lower-dimensional space [16]. This has shown to be able to use the hyperparameters of one robot to accelerate learning of the same skills by another kinematically similar robot.…”

Section: Related Workmentioning

confidence: 99%

“…The similarity of postures can be determined visually by the human teacher [15,16]. To this end, the robot is programmed to assume a pre-defined sequence of postures, such as the T-posture [36] and U-posture shown in Fig.…”

Section: Correspondencementioning

confidence: 99%

“…The focus of this paper is on how to adapt demonstrations gathered using vision-based systems, such that they are useful to the robot learner, and to demonstrate that policies can be derived from the adapted demonstrations. Several approaches have been proposed to deal with correspondence issues that arise in this setting, including optimization approaches based on a kinematic model of the robot learner [6,13,14], and data-driven approaches that attempt to model a mapping from the teacher space to the robot space [15][16][17]. Techniques that rely on a kinematic model of the robot learner are among the most widely used, and some general methods have been proposed [6,14].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

A Non-Linear Manifold Alignment Approach to Robot Learning from Demonstrations

Makondo¹,

Hiratsuka²,

Rosman³

et al. 2018

JRM

View full text Add to dashboard Cite

The number and variety of robots active in real-world environments are growing, as well as the skills they are expected to acquire, and to this end we present an approach for non-robotics-expert users to be able to easily teach a skill to a robot with potentially different, but unknown, kinematics from humans. This paper proposes a method that enables robots with unknown kinematics to learn skills from demonstrations. Our proposed method requires a motion trajectory obtained from human demonstrations via a vision-based system, which is then projected onto a corresponding human skeletal model. The kinematics mapping between the robot and the human model is learned by employing Local Procrustes Analysis, a manifold alignment technique which enables the transfer of the demonstrated trajectory from the human model to the robot. Finally, the transferred trajectory is encoded onto a parameterized motion skill, using Dynamic Movement Primitives, allowing it to be generalized to different situations. Experiments in simulation on the PR2 and Meka robots show that our method is able to correctly imitate various skills demonstrated by a human, and an analysis of the transfer of the acquired skills between the two robots is provided.

show abstract

Bypassing combinatorial explosions in equivalence structure extraction

Satoh

Yamakawa

2021

Knowl Inf Syst

View full text Add to dashboard Cite

Transfer learning of shared latent spaces between robots with similar kinematic structure

Cited by 26 publications

References 18 publications

Task space adaptation via the learning of gait controllers of magnetic soft millirobots

Task space adaptation via the learning of gait controllers of magnetic soft millirobots

A Non-Linear Manifold Alignment Approach to Robot Learning from Demonstrations

Bypassing combinatorial explosions in equivalence structure extraction

Contact Info

Product

Resources

About