Tactile SLAM: Real-time inference of shape and pose from planar pushing

Abstract

Tactile perception is central to robot manipulation in unstructured environments. However, it requires contact, and a mature implementation must infer object models while also accounting for the motion induced by the interaction. In this work, we present a method to estimate both object shape and pose in real-time from a stream of tactile measurements. This is applied towards tactile exploration of an unknown object by planar pushing. We consider this as an online SLAM problem with a nonparametric shape representation. Our formulation of tactile inference alternates between Gaussian process implicit surface regression and pose estimation on a factor graph. Through a combination of local Gaussian processes and fixedlag smoothing, we infer object shape and pose in real-time. We evaluate our system across different objects in both simulated and real-world planar pushing tasks.

Tactile Image-to-Image Disentanglement of Contact Geometry from Motion-Induced Shear

“…The combination of smooth servo control and local shape reconstruction was later used for full object reconstruction (Fig. 4, middle row), which remains an problem under active investigation in the field [23,24]. Our approach improves upon related work [5,6] that instead trained insensitivity to shear into the prediction network itself.…”

Tactile Pose Estimation and Policy Learning for Unknown Object Manipulation

“…A common assumption among these method is that the object is considered to be static. Suresh et al [47] considers the object motions and estimates both the object shape and its pose by alternating between Gaussian process implicit surface regression and factor graph. Similarly, Liang et al [37] takes account of the object motions while it is being held in a multi-fingered hand and localizes the object pose using derivative-free sampling-based optimization.…”

Active Multi-Object Exploration and Recognition via Tactile Whiskers

“…Active spatial exploration concerns the acquisition of the scene or object's spatial features using active exploration. The spatial exploration allows the agent to gather the information necessary to address challenges such as scene reconstruction [35], object recognition [36], pose estimation [13], and planning manipulation policies [37], which are stepstones in realistic applications leveraging robotics and machine perception.…”

“…In those cases, the contact observations cannot provide complete descriptions about the object contour due to inter-occlusion, which may lead to misrecognition. We envision this can be addressed by involving planar pushing techniques such as the method by Suresh et al [13] for contour tracing and object separation.…”

“…), multiple touches are required using small contact areas per touch, or fewer touches with larger contact areas. Enlarging the contact area generally requires larger contact forces to maintain the pressure needed, which can lead to object repositioning inadvertently [13]. All these factors affect the accuracy and reliability of the tactile feedback.…”