Category Archives: Robot Models

Embedding actual knowledge into Deep Learning to improve its reliability

Lutter M, Peters J., Combining physics and deep learning to learn continuous-time dynamics models, The International Journal of Robotics Research. 2023;42(3):83-107 DOI: 10.1177/02783649231169492.

Deep learning has been widely used within learning algorithms for robotics. One disadvantage of deep networks is that these networks are black-box representations. Therefore, the learned approximations ignore the existing knowledge of physics or robotics. Especially for learning dynamics models, these black-box models are not desirable as the underlying principles are well understood and the standard deep networks can learn dynamics that violate these principles. To learn dynamics models with deep networks that guarantee physically plausible dynamics, we introduce physics-inspired deep networks that combine first principles from physics with deep learning. We incorporate Lagrangian mechanics within the model learning such that all approximated models adhere to the laws of physics and conserve energy. Deep Lagrangian Networks (DeLaN) parametrize the system energy using two networks. The parameters are obtained by minimizing the squared residual of the Euler\u2013Lagrange differential equation. Therefore, the resulting model does not require specific knowledge of the individual system, is interpretable, and can be used as a forward, inverse, and energy model. Previously these properties were only obtained when using system identification techniques that require knowledge of the kinematic structure. We apply DeLaN to learning dynamics models and apply these models to control simulated and physical rigid body systems. The results show that the proposed approach obtains dynamics models that can be applied to physical systems for real-time control. Compared to standard deep networks, the physics-inspired models learn better models and capture the underlying structure of the dynamics.

Learning robot simulators

Grant W. Woodford, Mathys C. du Plessis, Bootstrapped Neuro-Simulation for complex robots, . Robotics and Autonomous Systems, Volume 136, 2021 DOI: 10.1016/j.robot.2020.103708.

Robotic simulators are often used to speed up the Evolutionary Robotics (ER) process. Most simulation approaches are based on physics modelling. However, physics-based simulators can become complex to develop and require prior knowledge of the robotic system. Robotics simulators can be constructed using Machine Learning techniques, such as Artificial Neural Networks (ANNs). ANN-based simulator development usually requires a lengthy behavioural data collection period before the simulator can be trained and used to evaluate controllers during the ER process. The Bootstrapped Neuro-Simulation (BNS) approach can be used to simultaneously collect behavioural data, train an ANN-based simulator and evolve controllers for a particular robotic problem. This paper investigates proposed improvements to the BNS approach and demonstrates the viability of the approach by optimising gait controllers for a Hexapod and Snake robot platform.

A so-called universal approach for modelling and controlling robots

Tarokh, M., A unified kinematics modeling, optimization and control of universal robots: from serial and parallel manipulators to walking, rolling and hybrid robots, . Auton Robot 44, 1233–1248 (2020) DOI: 10.1007/s10514-020-09929-6.

The paper develops a unified kinematics modeling, optimization and control that is applicable to a wide range of autonomous and non-autonomous robots. These include hybrid robots that combine two or more modes of operations, such as combination of walking and rolling, or rolling and manipulation, as well as parallel robots in various configurations. The equations of motion are derived in compact forms that embed an optimization criterion. These equations are used to obtain various useful forms of the robot kinematics such as recursive, body and limb-end kinematic forms. Using the modeling, actuation and control equations are derived that ensure traversing a desired path while maintaining balanced operations and tip-over avoidance. Various simulation results are provided for a hybrid rolling-walking robot, which demonstrate the capabilities and effectiveness of the developed methodologies.

Improving the realism of a simulator through deep learning

Allevato, A.D., Schaertl Short, E., Pryor, M. et al. , Iterative residual tuning for system identification and sim-to-real robot learning, . Auton Robot 44, 1167–1182 (2020) DOI: 10.1007/s10514-020-09925-w.

Robots are increasingly learning complex skills in simulation, increasing the need for realistic simulation environments. Existing techniques for approximating real-world physics with a simulation require extensive observation data and/or thousands of simulation samples. This paper presents iterative residual tuning (IRT), a deep learning system identification technique that modifies a simulator’s parameters to better match reality using minimal real-world observations. IRT learns to estimate the parameter difference between two parameterized models, allowing repeated iterations to converge on the true parameters similarly to gradient descent. In this paper, we develop and analyze IRT in depth, including its similarities and differences with gradient descent. Our IRT implementation, TuneNet, is pre-trained via supervised learning over an auto-generated simulated dataset. We show that TuneNet can perform rapid, efficient system identification even when the true parameter values lie well outside those in the network’s training data, and can also learn real-world parameter values from visual data. We apply TuneNet to a sim-to-real task transfer experiment, allowing a robot to perform a dynamic manipulation task with a new object after a single observation.

Automatic design of a robot to perform given tasks with an optimal configuration

Sehoon Ha et al., Computational Design of Robotic Devices From High-Level Motion Specifications, IEEE Transactions on Robotics, vol. 34, no. 5, DOI: 10.1109/TRO.2018.2830419.

We present a novel computational approach to design the robotic devices from high-level motion specifications. Our computational system uses a library of modular components—actuators, mounting brackets, and connectors—to define the space of possible robot designs. The process of creating a new robot begins with a set of input trajectories that specify how its end effectors and/or body should move. By searching through the combinatorial set of possible arrangements of modular components, our method generates a functional, as-simple-as-possible robotic device that is capable of tracking the input motion trajectories. To significantly improve the efficiency of this discrete optimization process, we propose a novel heuristic that guides the search for appropriate designs. Briefly, our heuristic function estimates how much an intermediate robot design needs to change before it becomes able to execute the target motion trajectories. We demonstrate the effectiveness of our computational design method by automatically creating a variety of robotic manipulators and legged robots. To generate these results, we define our own robotic kit that includes off-the-shelf actuators and 3-D printable connectors. We validate our results by fabricating two robotic devices designed with our method.

Calibrating a robotic manipulator through photogrammetry, and a nice state-of-the-art in the issue of robot calibration

Alexandre Filion, Ahmed Joubair, Antoine S. Tahan, Ilian A. Bonev, Robot calibration using a portable photogrammetry system, Robotics and Computer-Integrated Manufacturing, Volume 49, 2018, Pages 77-87, DOI: 10.1016/j.rcim.2017.05.004.

This work investigates the potential use of a commercially-available portablephotogrammetry system (the MaxSHOT 3D) in industrial robot calibration. To demonstrate the effectiveness of this system, we take the approach of comparing the device with a laser tracker (the FARO laser tracker) by calibrating an industrial robot, with each device in turn, then comparing the obtained robot position accuracy after calibration. As the use of a portablephotogrammetry system in robot calibration is uncommon, this paper presents how to proceed. It will cover the theory of robot calibration: the robot’s forward and inverse kinematics, the elasto-geometrical model of the robot, the generation and ultimate selection of robot configurations to be measured, and the parameter identification. Furthermore, an experimental comparison of the laser tracker and the MaxSHOT3D is described. The obtained results show that the FARO laser trackerION performs slightly better: The absolute positional accuracy obtained with the laser tracker is 0.365mm and 0.147mm for the maximum and the mean position errors, respectively. Nevertheless, the results obtained by using the MaxSHOT3D are almost as good as those obtained by using the laser tracker: 0.469mm and 0.197mm for the maximum and the mean position errors, respectively. Performances in distance accuracy, after calibration (i.e. maximum errors), are respectively 0.329mm and 0.352mm, for the laser tracker and the MaxSHOT 3D. However, as the validation measurements were acquired with the laser tracker, bias favors this device. Thus, we may conclude that the calibration performances of the two measurement devices are very similar.

Building probabilistic models of physical processes from their deterministic models and some experimental data, with guarantees on the degree of coincidence between the generated model and the real system

Konstantinos Karydis, Ioannis Poulakakis, Jianxin Sun, and Herbert G. Tanner, Probabilistically valid stochastic extensions of deterministic models for systems with uncertainty, The International Journal of Robotics Research September 2015 34: 1278-1295, first published on May 28, 2015. DOI: 10.1177/0278364915576336.

Models capable of capturing and reproducing the variability observed in experimental trials can be valuable for planning and control in the presence of uncertainty. This paper reports on a new data-driven methodology that extends deterministic models to a stochastic regime and offers probabilistic guarantees of model fidelity. From an acceptable deterministic model, a stochastic one is generated, capable of capturing and reproducing uncertain system–environment interactions at given levels of fidelity. The reported approach combines methodological elements from probabilistic model validation and randomized algorithms, to simultaneously quantify the fidelity of a model and tune the distribution of random parameters in the corresponding stochastic extension, in order to reproduce the variability observed experimentally in the physical process of interest. The approach can be applied to an array of physical processes, the models of which may come in different forms, including differential equations; we demonstrate this point by considering examples from the areas of miniature legged robots and aerial vehicles.

A nice review of the problem of kinematic modeling of wheeled mobile robots and a new approach that delays the use of coordinate frames

Alonzo Kelly and Neal Seegmiller, 2015, Recursive kinematic propagation for wheeled mobile robots, The International Journal of Robotics Research, 34: 288-313, DOI: 10.1177/0278364914551773.

The problem of wheeled mobile robot kinematics is formulated using the transport theorem of vector algebra. Doing so postpones the introduction of coordinates until after the expressions for the relevant Jacobians have been derived. This approach simplifies the derivation while also providing the solution to the general case in 3D, including motion over rolling terrain. Angular velocity remains explicit rather than encoded as the time derivative of a rotation matrix. The equations are derived and can be implemented recursively using a single equation that applies to all cases. Acceleration kinematics are uniquely derivable in reasonable effort. The recursive formulation also leads to efficient computer implementations that reflect the modularity of real mechanisms.