Dom, cars don't fly! -- Or do they? In-Air Vehicle Maneuver for High-Speed Off-Road Navigation

TL;DR

The paper tackles the challenge of high-speed off-road navigation where vehicles can become airborne, proposing a hybrid physics–learning model (phli) to predict angular accelerations and a fixed-horizon, sampling-based planner (Dom Planner) to reach a safe landing configuration within a short airborne window. The approach combines a forward kinodynamic model in $SE(3)$ with a data-driven acceleration predictor $g_oldsymbol{ heta}$ and a physics-based integrator $h_oldsymbol{\xi}$, enabling planning with throttle and steering controls alone. Key contributions include (1) a physics-informed learning framework for in-air maneuvering, (2) a fixed-horizon planner that converges to a feasible landing pose within the airborne interval, and (3) extensive indoor and outdoor validation on a 1/5-scale platform and gimbal dataset, demonstrating precise, timely landings despite disturbances. The results show that existing ground controls can be repurposed to perform controlled in-air maneuvers, enabling higher-speed off-road operation with safe, orientation-aware landings in realistic conditions.

Abstract

When pushing the speed limit for aggressive off-road navigation on uneven terrain, it is inevitable that vehicles may become airborne from time to time. During time-sensitive tasks, being able to fly over challenging terrain can also save time, instead of cautiously circumventing or slowly negotiating through. However, most off-road autonomy systems operate under the assumption that the vehicles are always on the ground and therefore limit operational speed. In this paper, we present a novel approach for in-air vehicle maneuver during high-speed off-road navigation. Based on a hybrid forward kinodynamic model using both physics principles and machine learning, our fixed-horizon, sampling-based motion planner ensures accurate vehicle landing poses and their derivatives within a short airborne time window using vehicle throttle and steering commands. We test our approach in extensive in-air experiments both indoors and outdoors, compare it against an error-driven control method, and demonstrate that precise and timely in-air vehicle maneuver is possible through existing ground vehicle controls.

Figures (4)

• Figure 1: In-air vehicle maneuvers are critical in ensuring safe vehicle landing during high-speed off-road navigation. Top: Precise and timely maneuvers prepare the robot to land with minimal impact. Bottom: Improper maneuvers cause the robot to land on its back, terminating mission execution and risking vehicle damage.
• Figure 2: Simplified Bicycle Model with Torques Acting on the Wheels and Chassis due to Wheel Acceleration and Steering.
• Figure 3: Two-Axis Gimbal for In-Air Dynamics.
• Figure 4: Qualitative Comparison of Roll and Pitch Angle Achieved by Dom Planner vs PID w.r.t. Goal Angle for Trajectory Tracking, Rapid State Change, Timed Goal Reaching, State Stability, and Outdoor Experiments: Dom Planner can closely track continuous Goal trajectory (left, top and middle), rapidly achieve discrete random Goal states (middle, top and middle), reach timed Goal in a timely manner (right, top and middle), keep stable at a constant Goal despite external disturbances (left, bottom), and maintain zero pitch and roll during outdoor flights (middle and right, bottom). PID suffers from poor performance due to a lack of knowledge about coupled vehicle dynamics among state dimensions and actuation limits.