Towards Semi-supervised Manifold Learning: UKR with Structural Hints

Abstract

Abstract. We explore generic mechanisms to introduce structural hints into the method of Unsupervised Kernel Regression (UKR) in order to learn representations of data sequences in a semi-supervised way. These new extensions are targeted at representing a dextrous manipulation task. We thus evaluate the effectiveness of the proposed mechanisms on appropriate toy data that mimic the characteristics of the aimed manipulation task and thereby provide means for a systematic evaluation.

Gait Reconstruction From Motion Artefact Corrupted Fabric-Embedded Sensors

“…As such, future work will investigate how information about the distribution of motion artefacts into the learning process (e.g., the heteroscedastic nature), can be used to better account for artefact generation. In addition unsupervised learning techniques that allow for prior knowledge about motions to be incorporated into manifold learning [35], will be investigated to enable not only noiseless measurement reconstruction, but the noiseless prediction of other unobserved points on the body (e.g., predicting end-effector information from sensors mounted on the upper arm). This will allow for greater applicability to systems that desire autonomous motion estimation e.g., wearable exoskeleton devices or intention-prediction in prosthetics.…”

“…Specifically, this is done by finding a latent space estimations * that minimises the (orthogonal) reconstruction error [35]:…”

Bio-inspired motion strategies for a bimanual manipulation task

“…To represent and learn complex motion trajectories from observation, we previously proposed Structured UKR manifolds [22], [23] which are based on Unsupervised Kernel Regression (UKR [24]), a recent approach to learn nonlinear continuous manifolds. UKR finds a low-dimensional (latent) representation X = {x 1 , x 2 , .…”

“…In the considered screwing task we use the cap radius as an parameter which defines the average opening of the hand. Structured UKR [23] extends the original cost function (3) with an additional term to penalize a disordering of data sequence elements along the time dimension. To learn periodic motions, a periodic kernel K(·) is used.…”

“…Maintaining a specific latent radius and progressing along the latent time dimension results in hand posture sequences, which resemble the trained turning motion for the given cap radius. Whereas this control scheme has been presented already in [22], [23], this is the first time that it has been applied to a real robot hand.…”