Distributed recurrent neural forward models with synaptic adaptation and CPG-based control for complex behaviors of...

Area of Science:

• Robotics and Artificial Intelligence
• Bio-inspired Engineering
• Computational Neuroscience

Background:

• Insects exhibit complex locomotive behaviors through biomechanics and neural control.
• Adaptation to environmental changes (uneven terrain, obstacles) is key to insect locomotion.
• Internal models are crucial for predictive capabilities in biological systems.

Purpose of the Study:

• To develop a bio-inspired walking system integrating biomechanics and neural mechanisms.
• To enable robots to perform complex locomotion tasks mimicking insect abilities.
• To investigate the efficacy of recurrent neural network-based internal models for robotic adaptation.

Main Methods:

• A bio-inspired robot combining body/leg structures with neural control was designed.
• Neural mechanisms included central pattern generators, recurrent neural networks (RNNs) for forward models, and leg-specific control.
• Simulations were used to test locomotion on varied terrains and conditions.

Main Results:

• The bio-inspired system demonstrated insect-like locomotion on undulated terrains and over obstacles.
• The robot successfully adapted to gap crossing and simulated leg damage.
• RNN-based online forward models outperformed previous state-of-the-art adaptive neuron models.

Conclusions:

• Combining biomechanics with advanced neural control, including RNN-based internal models, enables robust robotic locomotion.
• This approach allows for adaptive and complex behaviors in robots, similar to insects.
• The developed system represents a significant advancement in bio-inspired robotics and adaptive control.