End-Effector Force and Joint Torque Estimation of a 7-DoF Robotic Manipulator Using Deep Learning

When a mobile robotic manipulator interacts with other robots, people, or the environment in general, the end-effector forces need to be measured to assess if a task has been completed successfully. Traditionally used force or torque estimation methods are usually based on observers, which require knowledge of the robot dynamics. Contrary to this, our approach involves two methods based on deep neural networks: robot end-effector force estimation and joint torque estimation. These methods require no knowledge of robot dynamics and are computationally effective but require a force sensor under the robot base. Several different architectures were considered for the tasks, and the best ones were identified among those tested. First, the data for training the networks were obtained in simulation. The trained networks showed reasonably good performance, especially using the LSTM architecture (with a root mean squared error (RMSE) of 0.1533 N for end-effector force estimation and 0.5115 Nm for joint torque estimation). Afterward, data were collected on a real Franka Emika Panda robot and then used to train the same networks for joint torque estimation. The obtained results are slightly worse than in simulation (0.5115 Nm vs. 0.6189 Nm, according to the RMSE metric) but still reasonably good, showing the validity of the proposed approach.

Improvement of Torque Estimation for Series Viscoelastic Actuator Based on Dual Extended Kalman Filter

Adding damping such as viscoelastic element in series elastic actuators (SEA) can improve the force control bandwidth of the system and suppression of high frequency oscillations induced by the environment. Thanks to such advantages, series viscoelastic actuators (SVA) have recently gained increasing research interests from the community of robotic device design. Due to the inconvenience of mounting torque sensors, employing the viscoelastic elements to directly estimate the output torque is of great significance regarding the real-world applications of SVA. However, the nonlinearity and time-varying properties of viscoelastic materials would degrade the torque estimation accuracy. In such a case, it is paramount to simultaneously estimate the output torque state and viscoelastic model coefficients in order to enhance the torque estimation accuracy. To this end, this paper first completed the design of a rubber-based SVA device and used the Zenner linear viscoelastic model to model the viscoelastic element of the rubber. Subsequently, this paper proposed a dual extended Kalman filter- (DEFK) based torque estimation method to estimate the output torque and viscoelastic model coefficients simultaneously. The noisy observations of two Kalman filters were provided by motor current-based estimated torque. Moreover, the dynamic friction of harmonic drive of the designed SVA was modeled and compensated to enhance the reliability of current-based torque estimation. Finally, a number of experiments were carried out on SVA, and the experimental results confirmed the DEFK effectiveness of improving torque estimation accuracy compared to only-used rubber and only-used motor current torque estimation methods. Thus, the proposed method could be considered as an effective alternative approach of torque estimation for SVA.