Abstract |
One of the primary advantages of legged robots lies in their ability to navigate
through complex and unstructured environments, such as outdoor fields, sewers
and construction sites, which often feature a variety of challenging terrains. This
capability opens the door to employing legged robots in applications that might
pose risks to humans, including search and rescue missions, inspection tasks and
maintenance in critical infrastructure facilities.
In addition to the structural intricacies of these environments, they also present
dynamic challenges, with varying terrain friction being a prominent concern. Legged
robots frequently encounter the issue of partially or globally slippery terrains, which
can result from conditions like mud, wet surfaces, oil or ice. The slippage of any
leg relative to the supporting surface can introduce unpredictable and unmodeled
dynamics, potentially compromising trajectory tracking performance or even leading
to the robot’s instability from loss of contact with the supporting surface.
In such conditions, maintaining stability and precise control becomes paramount.
Ensuring that the robot follows desired trajectories with accuracy is not only
essential for its own safety but also critical for successfully executing dexterous
maneuvers in these challenging settings. Task space trajectory tracking plays
a central role in achieving these objectives, as it enables the robot to adapt to
the dynamic nature of its surroundings, react to unanticipated disturbances, and
minimize the risk of falls or instability. By focusing on accurate tracking of task
space trajectories, we aim to equip quadruped robots with the capability to operate
with confidence and reliability in the face of environmental uncertainties.
Driven by the challenges posed by agile maneuvers and locomotion in rough
and slippery terrains, we introduce an adaptive controller termed as the Body
Posture and Movement Controller (BPMC) designed specifically for such conditions.
BPMC comprises two key components: an adaptive trajectory tracking controller,
referred to as “Body Posture”, and an adaptive reaching-target controller that
initiates locomotion, called “Body Movement”.
The former, namely Body Posture controller, comprises a robust adaptive
trajectory tracking controller that consists of two prioritized layers of adaptation
aimed at maintaining stability during dynamic contact events of one or more
supporting legs. The main objective of the proposed adaptive controller is to induce
a robust reactive behaviour of a quadruped robot when it experiences unstable
contacts while executing a trajectory without sacrificing the spatial properties of
the task.
The Body Movement controller, serving as an adaptive reaching controller,
plays a pivotal role in initiating locomotion tasks and executing agile maneuvers,
particularly in challenging terrains marked by slipperiness and dynamic obstacles.
The core of the Body Movement controller lies in its initial layer, in which the
control effort is distributed among all stance legs, meaning all legs except the
swinging leg. The latter is accomplished by assigning an exceptionally high weight
to a specific leg, designated as the swinging leg. In that way, the swinging leg task
is attained while, at the same time, the robot keeps its stability and controllability
during locomotion.
On top of that, the Body Movement controller offers an additional layer that can
be activated at the user’s discretion, taking into account the probability of detecting
slip events. This extra layer draws inspiration from the approach used in the first
layer of the Body Posture controller. It dynamically adjusts the effort distribution
among all legs based on the slip probability of each foot. This multifaceted approach
not only introduces innovative concepts for agile movements but also ensures the
stability of the robot’s dynamic maneuvers. It represents a crucial step in advancing
the adaptability and robustness of the overall system.
The proposed methods constitute novel, lightweight analytical solutions that
assume no prior knowledge of the friction properties of the supporting surface. This
is accomplished by considering the slippage probability as extracted by our previous
work on contact state estimation in order to avoid non-controllable conditions.
Our experimental outcomes, stemming from both simulations and real-world
tests, highlight the approach’s effectiveness. It substantially enhances system
robustness, minimizing leg slippage while maintaining robot stability and control
even in challenging conditions. These advances mark significant milestones in
enhancing quadruped robot capabilities for diverse real-world scenarios.
|