We present a unified model-based and data-driven approach for quadrupedal planning control to achieve dynamic locomotion over uneven terrain. utilize on-board proprioceptive exteroceptive feedback map sensory information desired base velocity commands into footstep plans using reinforcement learning (RL) policy. This RL policy is trained in simulation wide range of procedurally generated terrains. When ran online, the system tracks motion controller. evaluate robustness our method variety...

In legged locomotion, the projection of robot's Center Mass (CoM) being inside convex hull contact points is a commonly accepted sufficient condition to achieve static balancing. However, some these configurations cannot be realized because joint-torques required sustain them would above their limits (actuation limits). this article, we rule out such and define feasible region, revisited support region that guarantees both global stability in sense tip-over slippage avoidance existence set...

Motion planning in multicontact scenarios has recently gathered interest within the legged robotics community, however actuator force/torque limits are rarely considered. We believe that these gain paramount importance when complexity of terrains to be traversed increases. build on previous research from field robotic grasping propose two new six-dimensional bounded polytopes named Actuation Wrench Polytope (AWP) and Feasible (FWP). define theAWP as set all wrenches a robot can generate...

To dynamically traverse challenging terrain, legged robots need to continually perceive and reason about upcoming features, adjust the locations timings of future footfalls leverage momentum strategically. We present a pipeline that enables flexibly-parametrized trajectories for perceptive dynamic quadruped locomotion be optimized in an online, receding-horizon manner. The initial guess passed optimizer affects computation needed achieve convergence quality solution. consider two methods...

We present a unified control framework that generates dynamic walking motions for biped and quadruped robots with online relative footstep optimization. The optimization is formulated as discrete-time Model Predictive Control problem which determines future locations. has hierarchical structure consisting of three layers: planner, trajectory generator whole-body controller. planner plans next position based on Linear Inverted Pendulum (LIP) model. Relative proposed to enable automatic...

The identification of inertial parameters is crucial to achieve high-performance model-based control legged robots. the legs are typically not altered during expeditions and therefore best identified offline. On other hand, trunk depend on modules mounted robot, like a motor provide hydraulic power, or different sets cameras for perception. This motivates use recursive approaches identify online mass position Center Mass (CoM) robot trunk, when payload change occurs. We propose two such...

We present a control architecture for real-time adaptation and tracking of trajectories generated using terrain-aware trajectory optimization solver. This approach enables us to circumvent the computationally exhaustive task online optimization, further introduces solution robust systems modeled with approximated dynamics. train policy deep reinforcement learning (RL) introduce additive deviations reference in order generate feedback-based system quadrupedal robot. this across multitude...

We propose a reactive locomotion strategy, called height reflex, that is useful to address big elevation changes in the terrain (e.g. when quadruped robot has step down from high platform). In these cases swing leg can lose mobility creating issues subsequent steps. The reflex foot trajectory replanning strategy redistributes motion (in smart way) stance legs "lower" whole trunk and aid foothold searching motion. To spread we exploit massless link model of with virtual dampers at joints,...

Simplified models are useful to increase the computational efficiency of a motion planning algorithm, but their lack accuracy have be managed. We propose two feasibility constraints included in Single Rigid Body Dynamics-based trajectory optimizer order obtain robust motions challenging terrain. The first one finds an approximate relationship between joint-torque limits and admissible contact forces, without requiring joint positions. second proposes leg model prevent collision with...

Robotic manipulation of fresh fruits and vegetables, including the grasping multiple loose items, has a strong industrial need but it still is challenging task for robotic manipulation. This paper outlines distinctive strategies used by humans to pick vegetables with aim better adopt them diverse items. In this work we present first version setup designed different single or featuring multi-fingered compliant gripper. We analyse human from perspective Key Performance Indicators (KPIs) in...

Developing feasible body trajectories for legged systems on arbitrary terrains is a challenging task. In this paper, we present paradigm that allows to design Center of Mass (CoM) and in an efficient manner. our previous work [1], introduced the notion 2D region, where static balance satisfaction joint torque limits were guaranteed, whenever projection CoM lied inside proposed admissible region. propose general formulation improved region guarantees dynamic alongside both joint-torque...

This manuscript presents a method to calculate and analyze mechanical shock of multi-rigid body system, based on the revised concept center percussion newly derived variable called radius percussion. The objective is improve mechanism’s robustness against shocks that are caused by certain impacts, such as those experienced legged robots from landing jump or making step. In practice, it can be used for placement shock-sensitive components in robots, inertial measurement units cameras,...

The motivation of our current research is to devise motion planners for legged locomotion that are able exploit the robot's actuation capabilities. This means, when possible, minimize joint torques or propel as much admissible required. For this reason we define two new 6 dimensional bounded polytopes name Actuation-consistent Wrench Polytope (AWP) and Feasible (FWP). These objects turn out be very useful in planning definition constraints on accelerations Center Mass robot respect friction...

The efficient evaluation the dynamic stability of legged robots on non-coplanar terrains is important when developing motion planning and control policies. inference time this measure has a strong influence how fast robot can react to unexpected events, plan its future footsteps or body trajectory. Existing approaches suitable for real-time decision making are either limited flat ground quasi-static locomotion. Furthermore, joint-space feasibility constraints usually not considered in...

Simplified models are useful to increase the computational efficiency of a motion planning algorithm, but their lack accuracy have be managed. We propose two feasibility constraints included in Single Rigid Body Dynamicsbased trajectory optimizer order obtain robust motions challenging terrain. The first one finds an approximate relationship between joint-torque limits and admissible contact forces, without requiring joint positions. second proposes leg model prevent collision with...