Analysis and Comparison of Slip Detection Methods for Quadruped Robots

Slippage is a problem that legged robots must deal with in several scenarios, and although there are methods to detect it, there is a lack of comparison between them. This work compares and analyzes a group of different slip detection methods based on thresholds for foot velocity and/or acceleration during the contact phase. The methods were implemented in simulation with the quadruped robot Gol using a Model Predictive Control (MPC) system for trajectory planning and a Whole Body Controller (WBC) to compute the actuation torques. The simulations are performed in two scenarios: a flat terrain with a constant friction coefficient of 0.8, and a flat terrain with a patch of slipping terrain with a friction coefficient of 0.1. The occurrence of slippage is measured by taking the real contact forces from the simulation engine and verifying if they are inside the correspondent friction cone of the terrain. The slip detection methods chosen for comparison employ a deterministic approach focused on proprioceptive sensors, and they have explicit descriptions of their implementation in the literature. This work also proposes a method that filters the slippage in the transition between the stance and swing phase, based on the relationship between the tangential and the orthogonal velocity of the foot at the moment of contact with the ground. The results highlight the relationship between the methods and evaluate their performance in terms of matches and mismatches with the ground truth.