Environmental force sensing helps robots traverse cluttered large obstacles

Robots can traverse sparse obstacles by sensing environmental geometry and avoiding contact with obstacles. However, for search and rescue in rubble, environmental monitoring through dense vegetation, and planetary exploration over Martian and lunar rocks, robots must traverse cluttered obstacles as large as themselves by physically interacting with them. Previous work discovered that the forest floor-dwelling discoid cockroach and a sensor-less minimalistic robot can traverse cluttered grass-like beam obstacles of various stiffness by transitioning across different locomotor modes. Yet the animal was better at traversal than the sensor-less robot, likely by sensing forces during obstacle interaction to control its locomotor transitions. Inspired by this, here we demonstrated in simulation that environmental force sensing helps robots traverse cluttered large obstacles. First, we developed a multi-body dynamics simulation and a physics model of the minimalistic robot interacting with beams to estimate beam stiffness from the sensed contact forces. Then, we developed a force feedback strategy for the robot to use the sensed beam stiffness to choose the locomotor mode with a lower mechanical energy cost. With feedforward pushing, the robot was stuck in front of stiff beams if it has a limited force capacity; without force limit, it traversed but suffered a high energy cost. Using obstacle avoidance, the robot traversed beams by avoiding beam contact regardless of beam stiffness, resulting in a high energy cost for flimsy beams. With force feedback, the robot determined beam stiffness, then traversed flimsy beams by pushing them over and stiff beams by rolling through the gap between them with a low energy cost. Stiffness estimation based on force sensing was accurate across varied body oscillation amplitude and frequency and position sensing uncertainty. Mechanical energy cost of traversal increased with sensorimotor delay. Future work should demonstrate cluttered large obstacle traversal using force feedback in a physical robot.

Learning quadrupedal locomotion on deformable terrain

Simulation-based reinforcement learning approaches are leading the next innovations in legged robot control. However, the resulting control policies are still not applicable on soft and deformable terrains, especially at high speed. The primary reason is that reinforcement learning approaches, in general, are not effective beyond the data distribution: The agent cannot perform well in environments that it has not experienced. To this end, we introduce a versatile and computationally efficient granular media model for reinforcement learning. Our model can be parameterized to represent diverse types of terrain from very soft beach sand to hard asphalt. In addition, we introduce an adaptive control architecture that can implicitly identify the terrain properties as the robot feels the terrain. The identified parameters are then used to boost the locomotion performance of the legged robot. We applied our techniques to the Raibo robot, a dynamic quadrupedal robot developed in-house. The trained networks demonstrated high-speed locomotion capabilities on deformable terrains: The robot was able to run on soft beach sand at 3.03 meters per second although the feet were completely buried in the sand during the stance phase. We also demonstrate its ability to generalize to different terrains by presenting running experiments on vinyl tile flooring, athletic track, grass, and a soft air mattress.

Virtual Energy Management for Physical Energy Savings in a Legged Robot Hopping on Granular Media

We discuss an active damping controller to reduce the energetic cost of a single step or jump of dynamic locomotion without changing the morphology of the robot. The active damping controller adds virtual damping to a virtual leg spring created by direct-drive motors through the robot’s leg linkage. The virtual damping added is proportional to the intrusion velocity of the robot’s foot, slowing the foot’s intrusion, and thus the rate at which energy is transferred to and dissipated by the ground. In this work, we use a combination of simulations and physical experiments in a controlled granular media bed with a single-leg robot to show that the active damping controller reduces the cost of transport compared with a naive compression-extension controller under various conditions.