A Multi-step Dynamics Modeling Framework For Autonomous Driving In Multiple Environments

• DOI: 10.48550/arxiv.2305.02241
• Publication Date: 2023-01-01
• Authors: Jason Gibson; Bogdan Vlahov; David P. Fan; Patrick Spieler; D. Pastor; Ali‐akbar Agha‐mo...; Evangelos A. Theo...
• Keywords: Odometry; Time horizon

Modeling dynamics is often the first step to making a vehicle autonomous. While on-road autonomous vehicles have been extensively studied, off-road pose many challenging modeling problems. An encounters highly complex and difficult-to-model terrain/vehicle interactions, as well having of its own. These complexities can create challenges for effective high-speed control planning. In this paper, we introduce framework multistep prediction that explicitly handles accumulation error remains scalable sampling-based controllers. Our method uses specially-initialized Long Short-Term Memory (LSTM) over limited time horizon learned component in hybrid model predict 4-person seating all-terrain (Polaris S4 1000 RZR) two distinct environments. By only LSTM fixed horizon, negate need long term stability challenge when training recurrent neural networks. flexible it requires odometry information labels. Through extensive experimentation, show our able millions possible trajectories real-time, with five seconds off road driving scenarios.