Skip to content
Snippets Groups Projects
Commit 6a11e214 authored by Amira Abdel-Rahman's avatar Amira Abdel-Rahman
Browse files

AI that grows

parent d7d68b09
Branches
No related tags found
No related merge requests found
Pipeline #5662 passed
...@@ -8,10 +8,11 @@ Research and development of workflows for the co-design reconfigurable AI softwa ...@@ -8,10 +8,11 @@ Research and development of workflows for the co-design reconfigurable AI softwa
- "focus on finding minimal architectures". - "focus on finding minimal architectures".
- "By deemphasizing learning of weight parameters, we encourage the agent instead to develop ever-growing networks that can encode acquired skills based on its interactions with the environment". - "By deemphasizing learning of weight parameters, we encourage the agent instead to develop ever-growing networks that can encode acquired skills based on its interactions with the environment".
![](./WANN_schematic.png)
![](./WANN_operators.png)
<img src="./square_biped.png" width="50%" /><img src="./square_biped.gif" width="50%" /> <img src="./WANN_schematic.png" width="70%" />
<img src="./WANN_operators.png" width="70%" />
<img src="./square_biped.png" width="35%" /><img src="./square_biped.gif" width="35%" />
## Case Study Implementation: Cart-Pole Swing Up ## Case Study Implementation: Cart-Pole Swing Up
...@@ -21,7 +22,7 @@ One of the most famous benchmarks of non-linear control, there is lots of approa ...@@ -21,7 +22,7 @@ One of the most famous benchmarks of non-linear control, there is lots of approa
WANN is interesting as it tries to get the simplest network that uses the input sensors (position, rotation and their derivatives) to the output (force). It focuses on learning **principles** and **not only tune weights**. WANN is interesting as it tries to get the simplest network that uses the input sensors (position, rotation and their derivatives) to the output (force). It focuses on learning **principles** and **not only tune weights**.
<img src="./swing_best.png" width="100%" /> <img src="./swing_best.png" width="75%" />
This is one of the outputs of the network and you can see because of it's simplicity it's not a black box and one can deduce the principles learnt [[1]](https://towardsdatascience.com/weight-agnostic-neural-networks-fce8120ee829): This is one of the outputs of the network and you can see because of it's simplicity it's not a black box and one can deduce the principles learnt [[1]](https://towardsdatascience.com/weight-agnostic-neural-networks-fce8120ee829):
- the position parameter is almost directly linked to the force, there is only an inverter which means that if the cart is on right or left of the center (+- x), it always try to **go to in the opposite direction** to the center. - the position parameter is almost directly linked to the force, there is only an inverter which means that if the cart is on right or left of the center (+- x), it always try to **go to in the opposite direction** to the center.
...@@ -30,9 +31,9 @@ This is one of the outputs of the network and you can see because of it's simpli ...@@ -30,9 +31,9 @@ This is one of the outputs of the network and you can see because of it's simpli
## Evolution and DICE Integration ## Evolution and DICE Integration
<img src="./wann_run.png" width="100%" /> <img src="./evol1.gif" width="50%" />
<img src="./evol1.gif" width="75%" /> <img src="./wann_run.png" width="50%" />
<img src="./quick.gif" width="100%" /> <img src="./quick.gif" width="100%" />
......
...@@ -12,23 +12,22 @@ Hardware Implementation: ...@@ -12,23 +12,22 @@ Hardware Implementation:
Using the [MetaVoxel simulation tool](https://gitlab.cba.mit.edu/amiraa/metavoxels): Using the [MetaVoxel simulation tool](https://gitlab.cba.mit.edu/amiraa/metavoxels):
![](./200114_simulation.PNG) <img src="./200114_simulation.PNG" width="75%" /><br/>
## Robotics Design ## Robotics Design
<img src="./dice_assembly.gif" width="100%" /><br/> <img src="./dice_assembly.gif" width="75%" /><br/>
## Assembly ## Assembly
- Option for different kinds of assembly (swarm assembly) - Option for different kinds of assembly (swarm assembly)
<img src="./assembly.gif" width="100%" /> <br/> <img src="./assembly.gif" width="75%" /> <br/>
## Control ## Control
<img src="./rover1.gif" width="75%" /> <img src="./rover1.gif" width="50%" />
<img src="./rover2.gif" width="75%" /><br/> <img src="./rover2.gif" width="50%" /><br/>
<img src="./r.gif" width="75%" /><br/> <img src="./r.gif" width="50%" /><br/>
<img src="./rr.gif" width="50%" /><br/>
<img src="./rr.gif" width="75%" /><br/>
## Optimization ## Optimization
......
0% Loading or .
You are about to add 0 people to the discussion. Proceed with caution.
Please register or to comment