Many applications require robots to move through complex 3-D terrain with large obstacles, such as self-driving, search and rescue, and extraterrestrial exploration. Although robots are already excellent at avoiding sparse obstacles, they still struggle in traversing cluttered large obstacles. To make progress, we need to better understand how to use and control the physical interaction with obstacles to traverse them. Forest floor-dwelling cockroaches can use physical interaction to transition between different locomotor modes to traverse flexible, grass-like beams of a large range of stiffness. Inspired by this, here we studied whether and how environmental force sensing helps robots make active adjustments to traverse cluttered large obstacles. We developed a physics model and a simulation of a minimalistic robot capable of sensing environmental forces during traversal of beam obstacles. Then, we developed a force-feedback control strategy, which estimated beam stiffness from the sensed contact force using the physics model. Then in simulation we used the estimated stiffness to control the robot to either stay in or transition to the more favorable locomotor modes to traverse. When beams were stiff, force sensing induced the robot to transition from a more costly pitch mode to a less costly roll mode, which helped the robot traverse with a higher success rate and less energy consumed. By contrast, if the robot simply pushed forward or always avoided obstacles, it would consume more energy, become stuck in front of beams, or even flip over. When the beams were flimsy, force sensing guided the robot to simply push across the beams. In addition, we demonstrated the robustness of beam stiffness estimation against body oscillations, randomness in oscillation, and uncertainty in position sensing. We also found that a shorter sensorimotor delay reduced energy cost of traversal.

翻译：许多应用要求机器人在具有大型障碍物的复杂三维地形中移动，例如自动驾驶、搜索救援和地外探索。尽管机器人已擅长避开稀疏障碍物，但在穿越杂乱的大型障碍物时仍面临挑战。为取得进展，我们需要更深入地理解如何利用和控制与障碍物的物理交互来实现穿越。栖息于森林地表的蟑螂能通过物理交互在不同运动模式间切换，从而穿越具有广泛刚度范围的可弯曲草状横梁。受此启发，我们研究了环境力感知是否以及如何帮助机器人通过主动调整穿越杂乱的大型障碍物。我们开发了一个物理模型和仿真系统，模拟了一种能在穿越横梁障碍物时感知环境力的极简机器人。随后，我们提出了一种力反馈控制策略，利用物理模型从感知的接触力中估计横梁刚度。在仿真中，我们使用估计的刚度控制机器人保持或切换到更有利的运动模式进行穿越。当横梁刚度较大时，力感知促使机器人从较高能耗的俯仰模式切换到较低能耗的滚动模式，从而以更高的成功率和更低的能耗完成穿越。相比之下，若机器人仅向前推进或始终避开障碍物，则会消耗更多能量，卡在横梁前甚至翻倒。当横梁柔软时，力感知引导机器人直接推越横梁。此外，我们验证了横梁刚度估计对机体振荡、振荡随机性以及位置感知不确定性具有鲁棒性。我们还发现，缩短感觉运动延迟可降低穿越过程中的能量消耗。