RBE Colloquium Series Presents
Towards autonomy of soft robotics snakes: Real-time simulation
Presenter: Renato G. Gasoto and Xuan Liu
Advisor: Jie Fu and Cagdas Onal
While soft robots provide numerous advantages over rigid-body systems, such as compliance and tolerance to collisions, they are challenging to control due to the infinitude of degrees of freedom, and difficult-to-model dynamics. Furthermore, learning-based control systems require a large amount of data to converge into a feasible solution, which would be too costly to implement directly in the real robot. By combining the Finite Elements Method with Lagrangian mechanics, we provide a physics simulator model that enables the use of soft actuators modeled as constraints. To achieve real-time simulation results, we leverage the parallel computing capabilities of GPUs.
Such high-fidelity simulator gives us a chance to study the dynamics and to develop learning-based control algorithms on the soft robot more efficiently. So far, a bio-inspired framework combining the central pattern generator (CPG) model and a reinforcement learning algorithm based on policy gradient method has been implemented in the stimulating environment. The work aims to realize efficient and adaptive locomotion control on a soft snake robot through learning the policy on operating the CPG activation signal. The optimal policy is learned from rewarded or penalized experience in thousands of trials driving the snake towards the goals. The experiment has helped us discover some useful and interesting patterns about the snake movement between neuroscience and engineering design.
Mixed-Granularity Human-Swarm Interaction
Presenter: Jayam Patel
Advisor: Carlo Pinciroli
We present an augmented reality human-swarm interface that combines two modalities of interaction: environment-oriented and robot-oriented. The environment-oriented modality allows the user to modify the environment (either virtual or physical) to indicate a goal to attain for the robot swarm. The robot-oriented modality makes it possible to select individual robots to reassign them to other tasks to increase performance or remedy failures. Previous research has concluded that environment-oriented interaction might prove more difficult to grasp for untrained users. In this paper, we report a user study which indicates that, at least in collective transport, environment-oriented interaction is more effective than purely robot-oriented interaction, and that the two combined achieve remarkable efficacy.