Skip to content
Snippets Groups Projects
Commit 56fce525 authored by Aritz's avatar Aritz
Browse files

Merge branch 'master' of https://git.code.tecnalia.com/aritz.martinez/a-mfea-rl

parents 34a06479 da126e27
Branches
No related tags found
No related merge requests found
Hide whitespace changes
Inline Side-by-side
Showing
with 56 additions and 57 deletions
README.md
+ 56
− 57
View file @ 56fce525
......@@ -10,10 +10,10 @@ In the framework, a reformulation of the well-known MFEA/MFEA-II algorithms is i
2. **Adapted crossover operator**: the crossover operator must support the previous aspects by preventing neural models from exchanging irrelevant information.
3. **Layer-based Transfer Learning**: unlike in traditional means to implement Transfer Learning, the number of layers to be transferred between models evolved for different tasks is autonomously decided by A-MFEA-RL during the search.
The code works on top of ![Metaworld-v1](https://github.com/rlworkgroup/metaworld). The experimentation carried out considers three scenarios; *TOY*, *MT-10/MT-10-R* and *MT-50/MT-50-R* (Results included in [Results](#results) Section ), *R* denotes random initialized episodes as in the next image:
The code works on top of [Metaworld-v1](https://github.com/rlworkgroup/metaworld). The experimentation carried out considers three scenarios; *TOY*, *MT-10/MT-10-R* and *MT-50/MT-50-R* (Results included in [Results](#results) Section ), *R* denotes random initialized episodes as in the next image:
<h3>MT-10-R results
<img src="/uploads/16013a3d9ec2050cca0e0616189bea87/out.gif" width="80%" /> </h3>
<img src="/uploads/c743460c60cddf1bb099ecae3ea6365d/MT10.gif" width="80%" /> </h3>
# Running the experimentation
......@@ -23,7 +23,7 @@ conda env create -f mujoco36.yml
conda activate mujoco36
```
A-MFEA-RL depends on Metaworld and ![MuJoco](https://github.com/openai/mujoco-py) (license required). To install Metaworld please follow the instructions in the ![official GitHub](https://github.com/rlworkgroup/metaworld) or run:
A-MFEA-RL depends on Metaworld and [MuJoco](https://github.com/openai/mujoco-py) (license required). To install Metaworld please follow the instructions in the [official GitHub](https://github.com/rlworkgroup/metaworld) or run:
```bash
pip install git+https://github.com/rlworkgroup/metaworld.git@master#egg=metaworld
......@@ -40,57 +40,56 @@ python3 exp.py -exp INT -t INT -p STR
* `-p`: STRING. Name of the folder under `summary` where results are saved.
# Results
| | | MT-10 | | | MT-10-R | | | MT-50 | | | MT-50-R | |
|-|:-:|:-:|:-:|:-:|:-:|:-:|:-:|:-:|:-:|:-:|:-:|:-:|
| **Environment name (complexity)** | **A** | **B** | **C** | **A** | **B** | **C** | **A** | **B** | **C** | **A** | **B** | **C** |
| assembly (H) | - | - | - | - | - | - | 0 | 0 | 0 | 0 | 0 | 0 |
| basketball (H) | - | - | - | - | - | - | 0 | 0 | 0 | 22 | 33 | 0 |
| bin-picking (H) | - | - | - | - | - | - | 0 | 0 | 0 | 0 | 0 | 11 |
| box-close (H) | - | - | - | - | - | - | 44 | 44 | 0 | 22 | 33 | 0 |
| button-press-topdown (M) | 100 | 100 | 100 | 100 | 89 | 91 | 100 | 100 | 100 | 100 | 100 | 97 |
| button-press-topdown-wall (H) | - | - | - | - | - | - | 67 | 78 | 100 | 67 | 100 | 100 |
| button-press (M) | - | - | - | - | - | - | 44 | 67 | 100 | 44 | 55 | 100 |
| button-press-wall (H) | - | - | - | - | - | - | 100 | 100 | 100 | 100 | 100 | 98 |
| coffee-button (H) | - | - | - | - | - | - | 44 | 78 | 100 | 56 | 89 | 100 |
| coffee-pull (M) | - | - | - | - | - | - | 78 | 100 | 0 | 100 | 100 | 70 |
| coffee-push (M) | - | - | - | - | - | - | 78 | 89 | 100 | 89 | 89 | 40 |
| dial-turn (H) | - | - | - | - | - | - | 100 | 100 | 100 | 100 | 100 | 99 |
| disassemble (H) | - | - | - | - | - | - | 0 | 0 | 0 | 0 | 0 | 0 |
| door-close (H) | - | - | - | - | - | - | 78 | 56 | 100 | 78 | 55 | 100 |
| door-lock (H) | - | - | - | - | - | - | 89 | 100 | 100 | 89 | 89 | 100 |
| door-open (H) | 100 | 33 | 100 | 100 | 100 | 100 | 78 | 67 | 100 | 67 | 67 | 100 |
| door-unlock (M) | - | - | - | - | - | - | 78 | 89 | 100 | 89 | 100 | 100 |
| drawer-close (H) | 100 | 100 | 100 | 100 | 100 | 100 | 79 | 89 | 100 | 67 | 78 | 100 |
| drawer-open (H) | 0 | 33 | 100 | 33 | 0 | 99 | 22 | 33 | 100 | 22 | 44 | 98 |
| faucet-close (M) | - | - | - | - | - | - | 100 | 67 | 100 | 78 | 44 | 81 |
| faucet-open (M) | - | - | - | - | - | - | 89 | 89 | 100 | 89 | 67 | 91 |
| hammer (H) | - | - | - | - | - | - | 33 | 56 | 100 | 11 | 67 | 100 |
| hand-insert (M) | - | - | - | - | - | - | 100 | 100 | 100 | 100 | 100 | 100 |
| handle-press-side (H) | - | - | - | - | - | - | 0 | 11 | 100 | 100 | 33 | 40 |
| handle-press (H) | - | - | - | - | - | - | 89 | 78 | 60 | 100 | 78 | 35 |
| handle-pull-side (H) | - | - | - | - | - | - | 56 | 67 | 0 | 56 | 89 | 0 |
| handle-pull (H) | - | - | - | - | - | - | 89 | 100 | 0 | 78 | 100 | 0 |
| lever-pull (M) | - | - | - | - | - | - | 0 | 0 | 0 | 0 | 0 | 0 |
| peg-insert-side (H) | 67 | 33 | 0 | 56 | 56 | 0 | 0 | 22 | 0 | 44 | 33 | 0 |
| peg-unplug-side (H) | - | - | - | - | - | - | 100 | 100 | 0 | 100 | 100 | 0 |
| pick-out-of-hole (H) | - | - | - | - | - | - | 0 | 0 | 0 | 0 | 0 | 0 |
| pick-place (H) | 66 | 100 | 0 | 0 | 0 | 0 | 44 | 11 | 0 | 33 | 11 | 0 |
| pick-place-wall (H) | - | - | - | - | - | - | 44 | 33 | 0 | 33 | 0 | 10 |
| plate-slide-back-side (M) | - | - | - | - | - | - | 100 | 89 | 40 | 78 | 89 | 45 |
| plate-slide-back (M) | - | - | - | - | - | - | 67 | 89 | 100 | 89 | 100 | 58 |
| plate-slide-side (M) | - | - | - | - | - | - | 100 | 89 | 100 | 55 | 100 | 100 |
| plate-slide (M) | - | - | - | - | - | - | 33 | 100 | 100 | 78 | 78 | 77 |
| push-back (E) | - | - | - | - | - | - | 89 | 100 | 0 | 89 | 100 | 71 |
| push (E) | 100 | 100 | 100 | 78 | 67 | 59 | 44 | 89 | 100 | 78 | 33 | 47 |
| push-wall (M) | - | - | - | - | - | - | 56 | 33 | 100 | 55 | 44 | 47 |
| reach (E) | 100 | 100 | 100 | 100 | 100 | 91 | 100 | 100 | 100 | 100 | 100 | 98 |
| reach-wall (E) | - | - | - | - | - | - | 100 | 100 | 100 | 100 | 100 | 98 |
| shelf-place (H) | - | - | - | - | - | - | 0 | 0 | 0 | 44 | 55 | 0 |
| soccer (E) | - | - | - | - | - | - | 67 | 78 | 0 | 55 | 33 | 48 |
| stick-pull (H) | - | - | - | - | - | - | 11 | 33 | 0 | 11 | 44 | 79 |
| stick-push (H) | - | - | - | - | - | - | 0 | 0 | 0 | 11 | 0 | 100 |
| sweep-into (E) | - | - | - | - | - | - | 100 | 78 | 100 | 67 | 89 | 80 |
| sweep (E) | - | - | - | - | - | - | 100 | 89 | 100 | 100 | 67 | 74 |
| window-close (H) | 33 | 33 | 100 | 100 | 78 | 100 | 67 | 44 | 100 | 89 | 44 | 100 |
| window-open (H) | 67 | 100 | 100 | 78 | 89 | 99 | 11 | 67 | 100 | 44 | 78 | 93 |
| **Average success rate** | 73.3 | 73.2 | **80.0** | **74.5** | 67.9 | 73.9 | 57.3 | **62.0** | 60.0 | 61.5 | **62.1** | 59.7 |
| **Environment name (complexity)** | **MT-10** | **MT-10-R** | **MT-50** | **MT-50-R** |
|-----------------------------------|-----------|-------------|-----------|-------------|
| assembly (H) | - | - | 0 | 0 |
| basketball (H) | - | - | 0 | 0 |
| bin-picking (H) | - | - | 0 | 11 |
| box-close (H) | - | - | 0 | 0 |
| button-press-topdown (M) | 100 | 91 | 100 | 97 |
| button-press-topdown-wall (H) | - | - | 100 | 100 |
| button-press (M) | - | - | 100 | 100 |
| button-press-wall (H) | - | - | 100 | 98 |
| coffee-button (H) | - | - | 100 | 100 |
| coffee-pull (M) | - | - | 0 | 70 |
| coffee-push (M) | - | - | 100 | 40 |
| dial-turn (H) | - | - | 100 | 99 |
| disassemble (H) | - | - | 0 | 0 |
| door-close (H) | - | - | 100 | 100 |
| door-lock (H) | - | - | 100 | 100 |
| door-open (H) | 100 | 100 | 100 | 100 |
| door-unlock (M) | - | - | 100 | 100 |
| drawer-close (H) | 100 | 100 | 100 | 100 |
| drawer-open (H) | 100 | 99 | 100 | 98 |
| faucet-close (M) | - | - | 100 | 81 |
| faucet-open (M) | - | - | 100 | 91 |
| hammer (H) | - | - | 100 | 100 |
| hand-insert (M) | - | - | 100 | 100 |
| handle-press-side (H) | - | - | 100 | 40 |
| handle-press (H) | - | - | 60 | 35 |
| handle-pull-side (H) | - | - | 0 | 0 |
| handle-pull (H) | - | - | 0 | 0 |
| lever-pull (M) | - | - | 0 | 0 |
| peg-insert-side (H) | 0 | 0 | 0 | 0 |
| peg-unplug-side (H) | - | - | 0 | 0 |
| pick-out-of-hole (H) | - | - | 0 | 0 |
| pick-place (H) | 0 | 0 | 0 | 0 |
| pick-place-wall (H) | - | - | 0 | 10 |
| plate-slide-back-side (M) | - | - | 40 | 45 |
| plate-slide-back (M) | - | - | 100 | 58 |
| plate-slide-side (M) | - | - | 100 | 100 |
| plate-slide (M) | - | - | 100 | 77 |
| push-back (E) | - | - | 0 | 71 |
| push (E) | 100 | 59 | 100 | 47 |
| push-wall (M) | - | - | 100 | 47 |
| reach (E) | 100 | 91 | 100 | 98 |
| reach-wall (E) | - | - | 100 | 98 |
| shelf-place (H) | - | - | 0 | 0 |
| soccer (E) | - | - | 0 | 48 |
| stick-pull (H) | - | - | 0 | 79 |
| stick-push (H) | - | - | 0 | 100 |
| sweep-into (E) | - | - | 100 | 80 |
| sweep (E) | - | - | 100 | 74 |
| window-close (H) | 100 | 100 | 100 | 100 |
| window-open (H) | 100 | 99 | 100 | 93 |
| **Average success rate** | 80.0 | 73.9 | 60.0 | 59.7 |
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Please register or to comment