Calibration Method to Improve Transfer from Simulation to Quadruped Robots

Abstract

Using passive compliance in robotic locomotion has been seen as a cheap and straightforward way of increasing the performance in energy consumption and robustness. However, the control for such systems remains quite challenging when using traditional robotic techniques. The progress in machine learning opens a horizon of new possibilities in this direction but the training methods are generally too long and laborious to be conducted on a real robot platform. On the other hand, learning a control policy in simulation also raises a lot of complication in the transfer. In this paper, we designed a cheap quadruped robot and detail a calibration method to optimize a simulation model in order to facilitate the transfer of parametric motor primitives. We present results validating the transfer of Central Pattern Generators (CPG) learned in simulation to the robot which already give positive insights on the validity of this method.

Body Randomization Reduces the Sim-to-Real Gap for Compliant Quadruped Locomotion

“…The robot used for this paper is an update of the Tigrillo robot (Willems et al, 2017) as described by Urbain et al (2018) (Figure 1A). Tigrillo is a low-cost platform built with off-the-shelf components and a structure laser cut out of ABS.…”

“…The Tigrillo platform has a parametric model (Figure 1) that is simulated in the Neurorobotics platform (NRP) (Falotico et al, 2017), using Gazebo configured with ODE (Drumwright et al, 2010) physics engine. The model is calibrated using the calibration method detailed by Urbain et al (2018). This method is an automated procedure in which both the model and real robot are considered sensor-to-actuator transfer functions.…”

“…CMA-ES optimization minimizes the difference between the sensor recording from the robot and the model ( y ( t ) and y θ ( t ), respectively), by tuning the model parameters (trueθ^). Figure adapted from Urbain et al (2018).…”

“…Improving the simulator accuracy is then reduced to matching the transfer function of these filters. Urbain et al (2018) provides an automated and parametrized calibration method that improves simulation accuracy by treating both the physical robot and its parametrized model as black box dynamical systems. It optimizes the similarity between the transfer functions by matching their sensor response to a given actuation input.…”

“…Especially with compliant robots, the dynamics of the model may be different from the physical robot. Therefor, we use a calibration method that focuses on replicating the dynamics, as described in a previous paper Urbain et al (2018).…”