Table 1 Precision comparison with other works.
| Â | Method | Length (cm) | Actuation of each section | Section number | Error (mm) | Error/Length | Test data | ||
|---|---|---|---|---|---|---|---|---|---|
Control Method | Data input | ||||||||
Li (2021)45 | Closed-loop & Learning | Q-learning Pretrained with Simulator | Simulator and multi-iteration static points | 63.0 | 3-Dofs of pneumatics | 4 | 23 (200 s) | 3.7% | Points to points |
You (2017)46 | Â | Q-learning | Multi-iteration static points | 63.0 | 2-Dofs of pneumatics | 4 | 8 (2D) | 1.3% | |
Centurelli (2021)47 | Open-loop & Learning | Artificial Neural Network (ANN) | Static points | 20 | 3 DoFs of pneumatics | 1 | 10.6 | 5.3% | Trajectory tracking |
Trust Region Policy Optimization (TRPO) | Static points and Dynamic trajectories | 3.2 | 1.6% | ||||||
Satheeshbabu (2019)48 | Reinforce-ment Learning | Static points | 31 | 3-Dofs of pneumatics | 1 | < 29.8 | 9.6% | Points to points | |
Chen (2016)49 | K-nearest- Neighbors Regression (KNNR) | Dynamic trajectories | 6.3 | 2 DoFs of tendon-driven and 1 DoF of screw-driven | 1 | 2.15 | 3.4% | Trajectory tracking | |
Marchese (2014)50 | Closed-loop | PID control | Geometry and captured trajectories | 23.0 | 1 DoFs of pneumatics | 6 | 7.1 (2D) | 3.1% | Points to points |
Liu (2017)28 | Open-loop | Static model | Geometry and mechanics | 25.5 | 3-Dofs of tendon-driven | 1 | 3.8-9.5 | 1.5- 3.7% | Points to points |
Our work | Inverse static model based on non-constant strain kinematics | Geometry and mechanics | 87.7 | 3 DoFs of tendon-driven | 3 | 4.8% | 4.8% | Trajectory tracking | |
