Phase lag bounded velocity planning for high performance path tracking

Towards Lifelong Self-Supervision: A Deep Learning Direction for Robotics

“…is the damping ratio and ω n = √ KI is the natural frequency. It is assumed here that K, B, and I are all constants representing the proportional and derivative gains, and the scalar moment of inertia around the rotation axis respectively under a canonical spring-mass-damper model Hebert et al (2015). Closed-loop feedback controllers such as these have well understood proofs of stability.…”

“…Namely, in spirit of the flurry of work on autonomous vehicles, we emphasize that motion planners and path tracking procedures generally fall under this umbrella notion as well. In particular, path tracking in navigation at the most primitive sense, leverage a heading and longitudinal controller like those described by Hebert et al (2015). Potential field methods for path planning Ge and Cui (2002); Wang and Chirikjian (2000); Khatib (1985) like Harmonic function path planning Connolly and Grupen (1993) have shown success dating back over two decades of research.…”

“…Especially with recent advances in GPU parallelization for efficient relaxation and the logarithmic transformations to prevent diminishing gradients Wray et al (2016), these classical methods still remain powerful path generators. An elegant control theoretic relaxation-based method to velocity planning presented by Hebert et al (2015) is shown to be done in linear time of the path, resulting in minimal path deviations and maximal performance envelope. Many of these navigation solutions are fast and robust-so instead of replacing them completely with a learned approximation, we suggest treating resulting plans like motion primitives or inherent behaviors in the context of forming complex visuomotor hierarchies.…”