Systems and methods for rapid-prototyped robotic devices
Various embodiments for fast prototyping of morphologies and controllers related to locomotion for a robotic device are disclosed.
1. A robotic device, comprising:
a body, the body formed from a substrate comprised of a plurality of layers; and
a plurality of limbs formed from the substrate and arranged along the body, including:
a first limb in operable engagement along a first lateral side of the body, the first limb defining a first fin,
a second limb in operable engagement along a second lateral side of the body opposite the first lateral side, the second limb defining a second fin,
wherein the plurality of limbs are configured such that during operation a radial edge of the first fin and the second fin articulates with a ground surface,
wherein operation and orientation of the plurality of limbs relative to the body is configured based upon a sample-efficient reinforcement learning method that considers changes to the robotic device and environment and leverages low-dimensional nature and periodicity of locomotion gaits to learn control policies for the robotic device that adapt to the ground surface, and
wherein the body and the plurality of limbs are configured with optimal locomotion policies generated using a Group Factor Policy Search algorithm modified with a periodicity constraint, wherein the Group Factor Policy Search algorithm is modified with a periodic feature function that enforces the periodicity constraint onto the locomotion policy, and further incorporates dimensionality reduction and information about group structure.
2. The robotic device of claim 1 , wherein the substrate is formed with flexure-based hinges created through selective removal of rigid material along predetermined bend axes.
3. The robotic device of claim 1 , wherein the plurality of limbs include two rotational degrees of freedom relative to the body, such that during forward movement of the robotic device the first fin and the second fin move down and back into the body as the body moves up and forward.
4. The robotic device of claim 1 , wherein the first fin and the second fin are curved.
5. The robotic device of claim 1 , wherein the body and the plurality of limbs form a general turtle-shape configuration such that a back end of the body is tapered, and an apical portion of the body is shaped to elevate the body above the ground surface with an upturned apex.
6. The robotic device of claim 1 , wherein the first fin and the second fin are designed using a predictive kinematic model that analyzed differences in motion and fin shape and relative effects upon experimental trajectories.
7. The robotic device of claim 1 , further comprising:
a plurality of open-loop controllers; and
a motor, the plurality of open-loop controllers used to generate motor commands for the motor which incorporate sinusoidal functions offset by 180 degrees to accommodate sweeping of the first fin and the second fin during down strokes of the plurality of limbs.
8. The robotic device of claim 1 , wherein the plurality of limbs are rotated by 90 degrees relative to the body such that the robotic device defines a general L-shape resembling a humeral angle of an adult sea turtle fin during crawling.
9. The robotic device of claim 1 , wherein the first fin and the second fin are formed using a multi-layered laminate comprising a plurality of paper layers and an adhesive layer.
10. The robotic device of claim 1 , wherein the first fin and the second fin are formed using a multi-layered laminate comprising a plurality of paper layers, a plurality of adhesive layers, and a reinforcing fiberglass coating.
11. The robotic device of claim 1 , wherein the first fin and the second fin are configured specifically for the ground surface and are interchangeable.
12. The robotic device of claim 1 , wherein the substrate is formed as a single planar shape, and the body and the plurality of arms limbs are erected using predetermined cut-out portions of the substrate.
13. The robotic device of claim 1 , wherein the sample-efficient reinforcement learning method is implemented using at least one computing device during testing of the robotic device and considers predetermined modifications to the plurality of limbs.
14. The robotic device of claim 1 , wherein the substrate comprises a plurality of rigid layers sandwiched around a plurality of adhesive layers, the plurality of adhesive layers sandwiched around a polyester film.
15. The robotic device of claim 1 , wherein the Group Factor Policy Search algorithm modified with the periodicity constraint models a control dimension α i of the locomotion policy given a point at time t as:
a
i
=
∑
j
(
w
ij
_
+
m
ij
+
e
ij
)
sin
(
t
T
720
°
+
j
-
1
J
360
°
)
wherein the term
sin
(
t
T
720
°
+
j
-
1
J
360
°
)
is associated with the periodic feature function and wherein the term
sin
(
t
T
720
°
+
j
-
1
J
360
°
)
enforces the periodicity constraint on the control dimension of the locomotion policy for the point at time t.
16. A robotic device comprising:
a foldable body having a plurality of fins oriented along opposite lateral sides of the foldable body and operatively engaged to the foldable body, and
wherein the foldable body and the plurality of fins are configured with optimal locomotion policies generated using a Group Factor Policy Search algorithm modified with a periodicity constraint, wherein the Group Factor Policy Search algorithm is modified with a periodic feature function that enforces the periodicity constraint onto the locomotion policy and further incorporates dimensionality reduction and information about group structure.
17. The robotic device of claim 16 , wherein the foldable body and the plurality of fins are made from a multi-layer composite laminate material.
18. The robotic device of claim 16 , wherein the plurality of fins include a curved fin.
19. The robotic device of claim 16 , wherein the foldable body is made from a material that has pliable portions.
20. The robotic device of claim 16 , wherein the Group Factor Policy Search algorithm modified with the periodicity constraint models a control dimension α i of the locomotion policy given a point at time t as:
a
i
=
∑
j
(
w
ij
_
+
m
ij
+
e
ij
)
sin
(
t
T
720
°
+
j
-
1
J
360
°
)
wherein the term
sin
(
t
T
720
°
+
j
-
1
J
360
°
)
is associated with the periodic feature function and wherein the term
sin
(
t
T
720
°
+
j
-
1
J
360
°
)
enforces the periodicity constraint on the control dimension of the locomotion policy for the point at time t.