Training-Independent Validation of a Learning-Based Controller for Stunt Maneuvers
Training-Independent Validation of a Learning-Based Controller for Stunt Maneuvers
This research presents a method for validating learning-based vehicle controllers independent of their training conditions. The study introduces 'G-Maneuver', a controller achieving a 90.11% success rate in stunt maneuvers, and employs a simulation-to-reality-to-simulation loop to ensure safety during real-world trials with a 16.9% acceptance rate. By comparing kinematic indicators like heading change and sideslip between simulation and reality, the authors identify reliable metrics for safe deployment while noting discrepancies in force and roll indicators.
The paper proposes a training-independent validation framework for learning-based controllers, using a Sim2Real2Sim loop and automotive safety analysis to identify safe deployment indicators.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
cceptance criteria absent from the training objective. A learning-based control policy, G-Maneuver, with 90.11% success is tested for stunt-level maneuvers. A looser terminal-heading score ranked the worst setting as the best. Because real stunt tests are safety-critical, a simulation-to-reality-to-simulation (Sim2Real2Sim) loop uses the simulation to select safe zones for real-vehicle trials with 16.9% acceptance. A
Open source S3
e at https://youtu.be/dN1Oes8pNIo.
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org