NN-based teleoperation boosts remote vehicle control
A new neural network framework slashes latency and boosts precision in remote vehicle control.
Researchers from George Washington University (Ning Ding and Azim Eskandarian) have published a paper on arXiv proposing a neural network-based teleoperation framework for remote-controlled vehicles. The system tackles two critical bottlenecks in direct teleoperation: communication latency and the operator’s inability to perceive unmodeled environmental disturbances such as aerodynamic drag or nonlinear tire-road dynamics.
The framework integrates the Wave Variable (WV) approach to passively guarantee stability under stochastic delays, while an adaptive Radial Basis Function Network (RBFN) actively compensates for vehicle-specific uncertainties. Unlike prior WV-neural architectures designed for robotic arms, this system features decoupled adaptive laws tailored for vehicle longitudinal and lateral dynamics. Simulation benchmarks show the RBFN achieves superior robustness against disturbances while requiring drastically less execution time—orders of magnitude faster than MPC and NMPC—making it suitable for resource-constrained edge computing. Hardware-in-the-loop experiments using a 1/10th scale vehicle over a 4G network validate the system’s practical feasibility, safety, and robust trajectory tracking under real-world road uncertainties.
- Combines Wave Variable (WV) for stability under latency with Radial Basis Function Network (RBFN) for real-time disturbance compensation
- Hardware-in-the-loop tests on a 1/10th scale vehicle over 4G network show robust trajectory tracking
- Execution time is orders of magnitude faster than MPC/NMPC, enabling edge deployment
Why It Matters
Enables precise, low-latency remote control of vehicles in unstable or unpredictable environments, unlocking safer teleoperation for drones, mining robots, and emergency response vehicles.