Code accompanying the paper Vision-Based Online Key Point Estimation of Deformable Robots.
Estimation errors with respect to the corresponding robot's length
| Estimation Technique. | Number of Cameras | Robot Type | Robot Length (mm) | Tip Error | |
|---|---|---|---|---|---|
| VOKE (ours) | CNN | 2 | WaxCast arm | 335 | 0.3%±0.2% |
| VOKE (ours) | CNN | 2 | SoPrA | 270 | 0.5%±0.4% |
| VOKE (ours) | CNN | 2 | Soft fish | 115 | 0.6%±0.6% |
| Camarillo et al. | 2D point-cloud fit | 3 | Soft arm | 160 | 4.8% |
| Vandini et al. | Line feature detector | 1 | Soft arm | 260 | 2.8% |
| Pedari et al. | LED light placement | 2 | Soft arm | 468* | 4.5% |
| AlBeladi et al. | Edge detection & curve fit | 1 | Soft arm | 287 | 4.5%±3.1% |
* not provided, calculated based on their estimation data
The repo was written using Python 3.8 with conda on Ubuntu 20.04
For an easy start, you can download our processed dataset on three different types of soft robots from Google Drive.
To run with our example python code in python, specify --dataset_folder and --label_folder in config.py to the path where the preprocessed data is stored.