Biomedical Engineering Reference
In-Depth Information
INPUT LAYER
T
1
T
2
T
3
T
4
C
1
C
2
A
1
A
2
A
3
A
4
B
1
B
2
HIDDEN
LAYER
M
1
M
2
M
3
M
4
M
5
OUTPUT LAYER
(a)
(b)
7.5
7
6
5
4
3
2
1
0
−
1
−
2
6
5
4
3
2
1
0
Average
Best
6.5
6
5.5
5
4.5
−
−
1
−
2
6
5
4
3
2
1
0
−
6
−
4
−
2
0
2
4
6
−
6
−
4
−
2
0
2
4
6
6
5
4
3
2
1
0
−
1
−
2
4
3.5
3
1
−
2
−
0
5
10
15
20
25
30
35
40
45
50
−
6
−
4
−
2
0
2
4
6
−
6
−
4
−
2
0
2
4
6
Generations
(c) (d)
Figure 4.1 Evolving a controller for a fixed morphology. (a) The morphology of the machine contains four legs
actuated with eight motors, four ground touch sensors, four angle sensors, and two chemical sensors. (b) The
machine is controlled by a recurrent neural net whose inputs are connected to the sensors and whose outputs are
connected to the motors. (c) Evolutionary progress shows how the target misalignment error reduces over
generations. (d) White trails show the motion of the machine towards high concentration (darker area). Black
trail shows strack when the chemical sensors are turned off. (From Bongard, J. C. (2002) Evolved Sensor Fusion
and Dissociation in an Embodied Agent, Proceedings of the EPSRC/BBSRC International Workshop Biologically-
Inspired Robotics: The Legacy of W. Grey Walter. With permission.)
one successful controller in four different chemical concentration fields is shown in Figure 4.1d.
The white trails, which mark the progress of the center of mass of the robot over time, show clearly
how the robot moves towards high concentration.
But what is more striking about this experiment is that the robot learned to perform essentially
two tasks: to locomote and to change orientation towards the high concentration. When the
chemical sensors are disabled, the robot moves forward but not towards the chemical concentration
(see black trail in Figure 4.1d). This shows that the network evolved two
independent
functions:
locomotion and gradient tracking.
Can this process also work for a real (not simulated) legged robot? We recently tried evolving
controllers for a dynamical, legged robot (Zykov et al., 2004). The nine-legged machine is
composed of two Stewart platforms back to back. The platforms are powered by 12 pneumatic
linear actuators, with power coming from an onboard 4500 psi paintball canister. While most
robotic systems use position-controlled actuators whose exact extension can be set, pneumatic
actuators of the kind used here are force-controlled. Like biological muscle, the controller can
specify the force and duration of the actuation, but not the position. It is therefore a challenging
control problem. The controller architecture for this machine was an open-loop pattern generator
that determines when to open and close pneumatic valves. The on-off pattern was evolved;
Search WWH ::
Custom Search