We present FORGE, a method for sim-to-real transfer of force-aware manipulation policies in the presence of significant pose uncertainty. During simulation-based policy learning, FORGE combines a force threshold mechanism with a dynamics randomization scheme to enable robust transfer of the learned policies to the real robot. At deployment, FORGE policies, conditioned on a maximum allowable force, adaptively perform contact-rich tasks while avoiding aggressive and unsafe behaviour, regardless of the controller gains. Additionally, FORGE policies predict task success, enabling efficient termination and autonomous tuning of the force threshold. We show that FORGE can be used to learn a variety of robust contact-rich policies, including the forceful insertion of snap-fit connectors. We further demonstrate the multistage assembly of a planetary gear system, which requires success across three assembly tasks: nut threading, insertion, and gear meshing.
8 August 2018 in IEEE Robotics and Automation Letters
To properly evaluate the ability of robots to operate autonomously in the real world, it is necessary to develop methods for quantifying their self-righting capabilities. Here, we improve upon a sampling-based framework for evaluating self-righting capabilities that was previously validated in two dimensions. To apply this framework to realistic robots in three dimensions, we require algorithms capable of scaling to high-dimensional configuration spaces. Therefore, we introduce a novel adaptive sampling approach that biases queries toward the transitional states of the system, thus, identifying the critical transitions of the system using substantially fewer samples. To demonstrate this improvement, we compare our approach to results that were generated via the previous framework and were validated on hardware platforms. Finally, we apply our technique to a high-fidelity three-dimensional model of a US Navy bomb-defusing robot, which was too complex for the previous framework to analyze.
15 May 2024 in IEEE Robotics and Automation Letters
Suction-based robotic grippers are common in industrial applications due to their simplicity and robustness but struggle with geometric complexity. Grippers that can handle varied surfaces as easily as traditional suction grippers would be more effective. Here we show how a fractal structure allows suction-based grippers to increase conformability and expand approach angle range. We develop a model for how the design parameters affect the range of curvatures and payloads a gripper can hold, and show when joint braking can increase payload capacity. We build a prototype gripper and develop a compact vacuum brake system for it. We then perform a series of experiments that demonstrate how the fractal structure enables corner grasping and expands the approach angle range from ${\text{5}}^{\circ }$ to ${\text{45}}^{\circ }$ . We also show that adding brakes improves performance by over 30% at moderate contact angles. This letter demonstrates how fractal structures increase the utility of suction grippers through underactuation and locking.