Abstract:Underwater soft robots are challenging to model and control because of their high degrees of freedom and their intricate coupling with water. In this paper, we present a method that leverages the recent development in differentiable simulation coupled with a differentiable, analytical hydrodynamic model to assist with the modeling and control of an underwater soft robot. We apply this method to Starfish, a customized soft robot design that is easy to fabricate and intuitive to manipulate. Our method starts wit… Show more
“…However, this process requires intensive computation power and is very lengthy. New research proposes the use of differentiable simulation as a method of modeling soft robots and bridging the simulation-to-reality gap [21].…”
Modeling of soft robotics systems proves to be an extremely difficult task, due to the large deformation of the soft materials used to make such robots. Reliable and accurate models are necessary for the control task of these soft robots. In this paper, a data-driven approach using machine learning is presented to model the kinematics of Soft Pneumatic Actuators (SPAs). An Echo State Network (ESN) architecture is used to predict the SPA’s tip position in 3 axes. Initially, data from actual 3D printed SPAs is obtained to build a training dataset for the network. Irregular-intervals pressure inputs are used to drive the SPA in different actuation sequences. The network is then iteratively trained and optimized. The demonstrated method is shown to successfully model the complex non-linear behavior of the SPA, using only the control input without any feedback sensory data as additional input to the network. In addition, the ability of the network to estimate the kinematics of SPAs with different orientation angles θ is achieved. The ESN is compared to a Long Short-Term Memory (LSTM) network that is trained on the interpolated experimental data. Both networks are then tested on Finite Element Analysis (FEA) data for other θ angle SPAs not included in the training data. This methodology could offer a general approach to modeling SPAs with varying design parameters.
“…However, this process requires intensive computation power and is very lengthy. New research proposes the use of differentiable simulation as a method of modeling soft robots and bridging the simulation-to-reality gap [21].…”
Modeling of soft robotics systems proves to be an extremely difficult task, due to the large deformation of the soft materials used to make such robots. Reliable and accurate models are necessary for the control task of these soft robots. In this paper, a data-driven approach using machine learning is presented to model the kinematics of Soft Pneumatic Actuators (SPAs). An Echo State Network (ESN) architecture is used to predict the SPA’s tip position in 3 axes. Initially, data from actual 3D printed SPAs is obtained to build a training dataset for the network. Irregular-intervals pressure inputs are used to drive the SPA in different actuation sequences. The network is then iteratively trained and optimized. The demonstrated method is shown to successfully model the complex non-linear behavior of the SPA, using only the control input without any feedback sensory data as additional input to the network. In addition, the ability of the network to estimate the kinematics of SPAs with different orientation angles θ is achieved. The ESN is compared to a Long Short-Term Memory (LSTM) network that is trained on the interpolated experimental data. Both networks are then tested on Finite Element Analysis (FEA) data for other θ angle SPAs not included in the training data. This methodology could offer a general approach to modeling SPAs with varying design parameters.
“…However, exact gradients can be difficult to derive symbolically for complex optimization problems. Instead, recent works have turned to automatic differentiation using differentiable programming to automatically compute gradients in problems such as 3D shape optimization [18], aircraft design optimization [19], robot design optimization [20], [21], and machine learning [22].…”
Section: A Differentiable Simulation and Temporal Logicmentioning
Signal temporal logic (STL) provides a powerful, flexible framework for specifying complex autonomy tasks; however, existing methods for planning based on STL specifications have difficulty scaling to long-horizon tasks and are not robust to external disturbances. In this paper, we present an algorithm for finding robust plans that satisfy STL specifications. Our method alternates between local optimization and local falsification, using automatically differentiable temporal logic to iteratively optimize its plan in response to counterexamples found during the falsification process. We benchmark our counterexample-guided planning method against state-of-the-art planning methods on two long-horizon satellite rendezvous missions, showing that our method finds high-quality plans that satisfy STL specifications despite adversarial disturbances. We find that our method consistently finds plans that are robust to adversarial disturbances and requires less than half the time of competing methods. We provide an implementation of our planner at https://github.com/MIT-REALM/architect.
“…In the past this would require black-box optimisation or calculation of numerical gradients but can now be achieved using gradients calculated using autodiff libraries. Examples of real2sim systemID include the optimisation of a pendulum start state and parameters [13], soft robot properties [14] and frictional properties [15]. However, one of the limiting factors for systemID using a differentiable simulator is the simulation time which has to be on the order of seconds or less.…”
Section: B Differentiable Simulatorsmentioning
confidence: 99%
“…Extending upon their earlier work Heiden et al [17] demonstrated the transference of a quadrupedal walking policy from simulation to a real quadruped using a large number of optimal trajectories generated from simulation. Another sim2real investigation conducted using differentiable simulators is work by Du et al where the authors implemented a sim2real approach for a underactuated underwater robot [14].…”
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.