No abstract

No abstract

Bionic amphibious robots are the intersection of biology and robotics; they have the advantages of environmental adaptability and maneuverability. An amphibious robot that combines walking and swimming move modes inspired by a crab (Portunus) is presented in this article. The outstanding characteristic of the robot is that its environmental adaptability relies on the bionic multi-modal movement, which is based on two modular bionic swimming legs and six modular walking legs. We designed the biomimetic crab robot based on the biological observation results. The design, analysis, and simulation of its structure and motion parameters are introduced in this paper. The swimming propulsion capability and the walking performance are verified through indoor, pool, and seaside experiments. In conclusion, the designed bionic crab robot provides a platform with practical application capabilities in amphibious environment detection, concealed reconnaissance, and aquaculture.

Amphibious environments formed from sand and water present a formidable challenge to the running motion of field robots, as the mixing of granular media (GM) and water makes the force laws of robotic legs more complicated during robot running. To this end, we extended the granular resistive force theory (RFT) to saturated wet granular media, named saturated granular RFT (SGRFT), which can be suitable for saturated wet sand submerged in water. This method can extend RFT for dry GM to saturated wet granular media (SWGM) by using the method’s velocity and depth coefficient. The force laws of the robotic legs in dry GM and SWGM were tested, compared, and analyzed. The difference in force laws between the two kinds of media, from the sensitivity to speed (10 mm/s~50 mm/s) and depth (0~60 mm), was calculated. More than 70% of the prediction results of the horizontal resistive force using SGRFT have an error of less than 6%. The effectiveness of the SGRFT in legged robots is proved by simulation and testing of three kinds of legs. The difference in force laws when running is proved by the experiments of the robot equipped with the propeller-leg in dry GM and SWGM, which is vital for amphibious robots working in shoal environments (including dry GM and SWGM ground).

IntroductionThe value approximation bias is known to lead to suboptimal policies or catastrophic overestimation bias accumulation that prevent the agent from making the right decisions between exploration and exploitation. Algorithms have been proposed to mitigate the above contradiction. However, we still lack an understanding of how the value bias impact performance and a method for efficient exploration while keeping stable updates. This study aims to clarify the effect of the value bias and improve the reinforcement learning algorithms to enhance sample efficiency.MethodsThis study designs a simple episodic tabular MDP to research value underestimation and overestimation in actor-critic methods. This study proposes a unified framework called Realistic Actor-Critic (RAC), which employs Universal Value Function Approximators (UVFA) to simultaneously learn policies with different value confidence-bound with the same neural network, each with a different under overestimation trade-off.ResultsThis study highlights that agents could over-explore low-value states due to inflexible under-overestimation trade-off in the fixed hyperparameters setting, which is a particular form of the exploration-exploitation dilemma. And RAC performs directed exploration without over-exploration using the upper bounds while still avoiding overestimation using the lower bounds. Through carefully designed experiments, this study empirically verifies that RAC achieves 10x sample efficiency and 25% performance improvement compared to Soft Actor-Critic in the most challenging Humanoid environment. All the source codes are available at https://github.com/ihuhuhu/RAC.DiscussionThis research not only provides valuable insights for research on the exploration-exploitation trade-off by studying the frequency of policies access to low-value states under different value confidence-bounds guidance, but also proposes a new unified framework that can be combined with current actor-critic methods to improve sample efficiency in the continuous control domain.