258
C. Morrow et al.
between solid ground and something more unstable, such as concrete to sand
[3]. Biomimetic robots hope to bridge the gap between what traditional robots
are capable of and the adaptable nature that living organisms represent [1,2,11].
Biomimetic robots also serve a very important function in providing insight as
to how biology works. By creating robotics that replicate animal physiology,
we can test a variety of different hypotheses, such as how organic features help
organisms navigate through their environment [8,18], or how the nervous system
controls rhythmic behavior [12,16].
By developing a robot based on the human anatomy of the legs, pelvis, and
back, we can begin to develop a robot that could easily navigate through a variety
of different terrains and environments, while also learning about muscle coordination in humans that control smooth motions such as gait. This biomimetic
robot will also allow us to investigate how central pattern generators (CPGs)
produce and modulate gait [6,13]. We can also produce results that will give
more insight as to why organisms have biarticulate musculature [5]. In order to
create this platform for testing these hypotheses, we must first work towards creating a physical structure with high fidelity to human anatomy that can closely
mimic human movement patterns.
This work describes an iterative algorithm whose goal is to determine pneumatic artificial muscle (PAM) placement and orientation on a robot that result
in torque profiles about lower body joints that closely match the torque profiles
of a human. PAMs are a way of replicating muscle driven actuation that is seen
in humans [14,17]. The human model from which this work is based has 92 Hilltype muscle actuators controlling the lumbrosacral and leg joints [7]. The robot
in development has reduced this to 54 muscle actuators [2]. However, even with
the reduction in the number of planned muscles, it becomes a tedious and time
consuming task to position PAM attachment locations by hand to find locations
that best capture human torque capabilities and also represents a reasonably
constructed robot model. This algorithm speeds up the development and construction of a biomimetic bipedal robot by determining the optimal placement
of PAMs to best meet human capabilities.
2 Methods
The optimization algorithm was created in Matlab [15]. It begins with a calculation of the maximum torque that can be developed by human muscles about a
each joint of the lower body: lumbrosacral, hip, knee, ankle, subtalar, and MTP.
This torque calculation is determined by the muscle locations found in OpenSim
model Gait2392 [4], through the methods described by Bolen and Hunt [2]. The
algorithm then begins to systematically move muscle attachment points while
minimizing a cost function. Figure 1 shows images of the robot in development
as well as detailing the attachment points that are being optimized with this
work.
C. Morrow et al.
between solid ground and something more unstable, such as concrete to sand
[3]. Biomimetic robots hope to bridge the gap between what traditional robots
are capable of and the adaptable nature that living organisms represent [1,2,11].
Biomimetic robots also serve a very important function in providing insight as
to how biology works. By creating robotics that replicate animal physiology,
we can test a variety of different hypotheses, such as how organic features help
organisms navigate through their environment [8,18], or how the nervous system
controls rhythmic behavior [12,16].
By developing a robot based on the human anatomy of the legs, pelvis, and
back, we can begin to develop a robot that could easily navigate through a variety
of different terrains and environments, while also learning about muscle coordination in humans that control smooth motions such as gait. This biomimetic
robot will also allow us to investigate how central pattern generators (CPGs)
produce and modulate gait [6,13]. We can also produce results that will give
more insight as to why organisms have biarticulate musculature [5]. In order to
create this platform for testing these hypotheses, we must first work towards creating a physical structure with high fidelity to human anatomy that can closely
mimic human movement patterns.
This work describes an iterative algorithm whose goal is to determine pneumatic artificial muscle (PAM) placement and orientation on a robot that result
in torque profiles about lower body joints that closely match the torque profiles
of a human. PAMs are a way of replicating muscle driven actuation that is seen
in humans [14,17]. The human model from which this work is based has 92 Hilltype muscle actuators controlling the lumbrosacral and leg joints [7]. The robot
in development has reduced this to 54 muscle actuators [2]. However, even with
the reduction in the number of planned muscles, it becomes a tedious and time
consuming task to position PAM attachment locations by hand to find locations
that best capture human torque capabilities and also represents a reasonably
constructed robot model. This algorithm speeds up the development and construction of a biomimetic bipedal robot by determining the optimal placement
of PAMs to best meet human capabilities.
2 Methods
The optimization algorithm was created in Matlab [15]. It begins with a calculation of the maximum torque that can be developed by human muscles about a
each joint of the lower body: lumbrosacral, hip, knee, ankle, subtalar, and MTP.
This torque calculation is determined by the muscle locations found in OpenSim
model Gait2392 [4], through the methods described by Bolen and Hunt [2]. The
algorithm then begins to systematically move muscle attachment points while
minimizing a cost function. Figure 1 shows images of the robot in development
as well as detailing the attachment points that are being optimized with this
work.