Force Push: Robust Single-Point Pushing with Force Feedback

TL;DR

This work tackles robust planar pushing of unknown convex sliders using only force feedback from a single contact point. It proposes a task-space pushing controller that steers pushing direction based on force angle error and lateral path offset, without requiring a model of the slider, and extends it with contact-recovery, obstacle-admittance control, and force filtering. The approach is validated in simulation and hardware, showing convergence to the desired path for straight and curved trajectories and resilience to collisions with static obstacles, even when frictional and inertial parameters are varied. The results demonstrate that force-based pushing can effectively replace vision-driven pose tracking for single-point pushing, enabling robust operation in environments with uncertain object properties and limited sensing.

Abstract

We present a controller for quasistatic robotic planar pushing with single-point contact using only force feedback to sense the pushed object. We consider an omnidirectional mobile robot pushing an object (the "slider") along a given path, where the robot is equipped with a force-torque sensor to measure the force at the contact point with the slider. The geometric, inertial, and frictional parameters of the slider are not known to the controller, nor are measurements of the slider's pose. We assume that the robot can be localized so that the global position of the contact point is always known and that the approximate initial position of the slider is provided. Simulations and real-world experiments show that our controller yields pushes that are robust to a wide range of slider parameters and state perturbations along both straight and curved paths. Furthermore, we use an admittance controller to adjust the pushing velocity based on the measured force when the slider contacts obstacles like walls.

Figures (14)

• Figure 1: Our robot pushing a blue box across the floor using single-point contact. The contact force is measured by a force-torque sensor in the robot's wrist, but no other measurements of the object are provided. A video of our approach is available at http://tiny.cc/force-push.
• Figure 2: Example of our pushing controller with a square slider. The goal is to push the slider along the path $\bm{p}_d(s)$ by pushing with velocity $\bm{v}_p$ at the contact point $\bm{c}$. The resulting contact force $\bm{f}$ lies inside the friction cone (green). The pushing angle $\theta_p$ is proportional to the lateral offset $\Delta_c$ and difference between measured force angle and the desired path heading $\Delta_f=\theta_f-\theta_d$; all angles are measured with respect to the global fixed frame. In this example, the pushing velocity $\bm{v}_p$ will eventually rotate the slider so that the contact force points back toward the desired path. Depending on the contact friction coefficient $\mu_c$, the contact point may slip along the slider's edge over the course of a trajectory.
• Figure 3: Block diagram of the system. The components of our controller (in green) use measurements of the robot's pose $\bm{q}$ and contact force $\bm{f}$ to produce joint velocity inputs $\bm{u}$ that push the slider along a desired path $\bm{p}_d(s)$.
• Figure 4: Basic obstacle avoidance of the pusher. When the contact point $\bm{c}$ is within distance $\delta_{\min}$ of an obstacle, the direction of motion is adjusted to not move any closer to it.
• Figure 5: Examples of two sliders, located at position $\bm{r}$ and orientation $\phi$ in the global frame. The contact with the pusher is located at point $\bm{c}$, which is distance $\alpha$ along the slider's edge from a reference point. The contact force must lie in the friction cone at the contact point (shown in green).