Interactive Robot-Environment Self-Calibration via Compliant Exploratory Actions

TL;DR

This work proposes a self-calibration framework where robot-environmental spatial relations are automatically estimated through compliant exploratory actions by the robot itself, autonomously purely through interactive exploration of the environment’s geometries.

Abstract

Calibrating robots into their workspaces is crucial for manipulation tasks. Existing calibration techniques often rely on sensors external to the robot (cameras, laser scanners, etc.) or specialized tools. This reliance complicates the calibration process and increases the costs and time requirements. Furthermore, the associated setup and measurement procedures require significant human intervention, which makes them more challenging to operate. Using the built-in force-torque sensors, which are nowadays a default component in collaborative robots, this work proposes a self-calibration framework where robot-environmental spatial relations are automatically estimated through compliant exploratory actions by the robot itself. The self-calibration approach converges, verifies its own accuracy, and terminates upon completion, autonomously purely through interactive exploration of the environment's geometries. Extensive experiments validate the effectiveness of our self-calibration approach in accurately establishing the robot-environment spatial relationships without the need for additional sensing equipment or any human intervention.

Figures (6)

• Figure 1: Our self-calibration framework estimates the robot-environment spatial relationship via compliant exploratory actions. Visualized in green is the environment's pose as currently estimated by the robot, and in blue is the ground truth.
• Figure 2: The geometry of the robot's end-effector is approximated as a point cloud $\mathcal{P}^e$. The grid represents a voxelized cache SDF $\Phi_\mathcal{O}$ where each voxel's color corresponds to the distance to the surface boundary $\partial O$ of the environment. The left figure shows a non-contact scenario where the signed distance $d_t^{m,j}$ between the end-effector and the environment surface is positive; the right figure shows a penetration scenario where the signed distance is less than a threshold, $d_t^{m,j} < -\delta_P$.
• Figure 3: (a) An illustrative plot of a sliding action $U_t$. The blue dot is the reference contact location $r_t$ on the environment surface, and the black arrow represents the normal vector $\hat{n}_t$ at the contact location. The recorded observations for contact and non-contact are shown by the red and green dots, respectively. (b) Examples of convex segmentation for objects in Fig. \ref{['pic:objects']}. (c) Examples of a real robot executing two different sliding actions.
• Figure 4: Self-calibration performance is evaluated in simulation with 100,000 particles. The reported data illustrate the effects of: (a) SDF resolution, (b) end-effector point cloud size $L$, and (c) $\sigma_P$ (a particle weight evaluation parameter). The results, averaged over 5 experimental runs for each environmental object, showcase the final translational error (cm), with error bars representing the standard deviation.
• Figure 5: Environmental objects used in experiments: Baxter (real), table (sim/real), shelf (sim/real), beanbag (sim), nightstand (sim), winerack (sim). The green overlays are the objects' mesh models, displayed relative to the robot's base frame based on the robot's self-calibrated pose.