Hybrid Learning of Time-Series Inverse Dynamics Models for Locally Isotropic Robot Motion

Abstract

Applications of force control and motion planning often rely on an inverse dynamics model to represent the high-dimensional dynamic behavior of robots during motion. The widespread occurrence of low-velocity, small-scale, locally isotropic motion (LIMO) typically complicates the identification of appropriate models due to the exaggeration of dynamic effects and sensory perturbation caused by complex friction and phenomena of hysteresis, e.g., pertaining to joint elasticity. We propose a hybrid model learning base architecture combining a rigid body dynamics model identified by parametric regression and time-series neural network architectures based on multilayerperceptron, LSTM, and Transformer topologies. Further, we introduce a novel joint-wise rotational history encoding, reinforcing temporal information to effectively model dynamic hysteresis. The models are evaluated on a KUKA iiwa 14 during algorithmically generated locally isotropic movements. Together with the rotational encoding, the proposed architectures outperform stateof-the-art baselines by a magnitude of 10 3 yielding an RMSE of 0.14 Nm. Leveraging the hybrid structure and time-series encoding capabilities, our approach allows for accurate torque estimation, indicating its applicability in critically force-sensitive applications during motion sequences exceeding the capacity of conventional inverse dynamics models while retaining trainability in face of scarce data and explainability due to the employed physics model prior.

Artificial physics engine for real-time inverse dynamics of arm and hand movement

“…Yet, we achieve similar torque prediction performance (less than 0.15 Nm errors with high level of noise) as in these low-dimensional systems (Çallar and Böttger 2022). Furthermore, we propose this approach as the extension for the full musculoskeletal dynamics that details moment arm and muscle length relationships with posture (Smirnov et al 2021).…”

Artificial physics engine for real-time inverse dynamics of arm and hand movement

“…Previously, inverse dynamic transformations with ANNs were typically limited to several degrees of freedom for the control of robotic devices (e.g., Kuka arm robot, 5–7 DOFs) [ 38 , 39 ]. Yet, we achieve similar torque prediction performance (less than 0.15 Nm errors with a high level of noise) as in these low-dimensional systems [ 40 ]. Furthermore, we propose this approach as the extension for the full musculoskeletal dynamics that details moment arm and muscle length relationships with postures [ 3 ].…”

Velocity-aware spatial-temporal attention LSTM model for inverse dynamic model learning of manipulators

“…To further illustrate the progressiveness of the proposed method, the optimal combination network obtained above is compared with LSTM (Greff et al, 2016), GRU (Cho et al, 2014), RNN (Mukhopadhyay et al, 2019), MLP (Yilmaz et al, 2020), andTransformer (Çallar andBöttger, 2022) on three different data sets, as shown in Tables 4-6. Compared with the LSTM network, the training results on three datasets show that the proposed network has improved the average estimation accuracy of joint torque by 61.88%, 43.93%, and 71.13%, respectively.…”

“…It combines the non-linear modeling ability of the black-box method with the interpretability of the white-box method. At present, graybox methods rely more on the modeling accuracy of white-box methods (Çallar and Böttger, 2022;Reuss et al, 2022), but it has not yet reached the level where the advantages of both white-box and black-box methods can be fully utilized. The authors believe that compared with other fields, the problems of black-box methods in inverse dynamics modeling of manipulators have not been fully discussed.…”