Stable grasping under pose uncertainty using tactile feedback

Abstract

This paper deals with the problem of stable grasping under pose uncertainty. Our method utilizes tactile sensing data to estimate grasp stability and make necessary hand adjustments after an initial grasp is established. We first discuss a learning approach to estimating grasp stability based on tactile sensing data. This estimator can be used as an indicator to the stability of the current grasp during a grasping procedure. We then present a tactile experience based hand adjustment algorithm to synthesize a hand adjustment and optimize the hand pose to achieve a stable grasp. Experiments show that our method improves the grasping performance under pose uncertainty.

Appendix: simulating tactile sensors

Figure 19 shows an example of tactile sensor simulation on a Barrett hand. To simulate tactile sensors, the first step is to find a contact model to approximate the force distribution near each contact point. In the GraspIt! simulator, a robotic hand and a graspable object are both treated as rigid bodies. Thus, each contact detected by a collision detection system is initially modeled as a point contact. In the real world, however, the hand and the object in contact are actually deformable to some extent, resulting in an area in contact rather than a point. A point contact assumption then no longer holds reasonably. To simulate the contact region between the two bodies touching each other, we use a soft finger contact model as is developed in (Ciocarlie et al. 2007b). This model takes into account the local geometry and structure of the objects in contact and captures frictional effects such as coupling between tangential force and frictional torque. It locally approximates the surfaces of the two touching bodies as

$$\begin{aligned} z_i = A_ix^2+B_iy^2+C_ixy, \quad i\in \{1,2\} \end{aligned}$$

(10)

where the local contact coordinate system has its origin at the center of the contact and the $z$ axis aligned with the normal of the contact. The subscript $i$ distinguishes the contacting bodies from each other.

Fig. 19

Tactile sensor simulation. a A robotic grasp of a Barrett hand on a mug. b The simulated sensor responses on the hand. c–e The tactile responses on finger F1, F2, F3 respectively. f The tactile readings on the palm. The tactile readings go from black (no response) to pink (saturation) (Color figure online)

Based on this approximation, we can deduce the separation $h$ between two surfaces in the form of

$$\begin{aligned} h = \frac{1}{2R^{\prime }}x^2 + \frac{1}{2R^{\prime \prime }}y^2 \end{aligned}$$

(11)

where $R^{\prime }$ and $R^{\prime \prime }$ are the relative radii of curvature of the objects in contact, depending only on their local geometry.

After a contact region is determined, we consider how the forces are formed within the contact region so that the response of the corresponding tactile sensor cells can be analyzed and evaluated. To express the pressure distribution inside a contact region using non-planar models that take into account the local geometry of the objects involved, we choose a Hertzian model as used in (Ciocarlie et al. 2007b).

In this model, the ratio of frictional torque to contact load which is used to compute the eccentricity parameter of the friction ellipsoid can be obtained from

$$\begin{aligned} \frac{max(\tau _n)}{P}=\frac{3\pi }{16}\mu \sqrt{ab} \end{aligned}$$

(12)

where $\mu$ is the frictional coefficient, $\tau _n$ is a frictional moment about the contact normal, $P$ is the contact load, and $a$ and $b$ are the lengths of the semi-axes.

Based on the soft finger contact model, we compute the contact region for a hand-object contact as well as the pressure distribution within the contact region. Since a tactile sensor cell performs as an atomic sensing unit, we discretize the soft finger contact region so that we can accumulate the total forces within each discrete part and use this to compute the forces sensed on each corresponding tactile sensor cell. We summarize the procedure to generate the tactile feedback of a robotic grasp in Algorithm 4.