Modular soft robots (MSRs) exhibit greater potential for sophisticated tasks compared with single-module robots. However, the modular structure incurs the complexity of accurate control and necessitates a control strategy specifically for modular robots. In this article, we introduce a data collection strategy tailored for MSR and a bidirectional long short-term memory (biLSTM) configuration controller capable of adapting to varying module numbers. Simulation cable-driven robots and real pneumatic robots have been included in experiments to validate the proposed approaches. Experimental results have demonstrated that MSRs can explore a larger space, thanks to our data collection method, and our controller can be leveraged despite an increase or decrease in module number. By leveraging the biLSTM, we aim to mimic the physical structure of MSRs, allowing the controller to adapt to module number change. Future work may include a planning method that bridges the task, configuration, and actuation spaces. We may also integrate online components into this controller.
Keywords: bidirectional LSTM; configuration control; data-driven control; modular soft robot.