Demonstrating HOUND: A Low-cost Research Platform for High-speed Off-road Underactuated Nonholonomic Driving

TL;DR

The paper tackles the challenge of enabling rapid, safe research in high-speed off-road autonomy by introducing HOUND, a low-cost 1/10-scale platform that combines open-source software with a hardware stack designed for rugged outdoor use. It integrates a GPU-accelerated perception pipeline with a single-track dynamics model and a model-predictive control framework, and it couples SITL/HITL testing in BeamNG to safely push performance limits. A core contribution is the rollover prevention system (RPS), comprising a static limiter and a feedback-based LQR controller that relies only on wheel-speed and IMU data, significantly reducing rollover risk in high-speed off-road trials. Real-world testing across 50 km on diverse terrains demonstrates the platform’s longevity and robustness, validating the safety advantages of RPS and the practicality of a low-cost, portable autonomy stack for broad research use and data sharing.

Abstract

Off-road autonomy, crucial for applications such as search-and-rescue, agriculture, and planetary exploration, poses unique problems due to challenging terrains, as well as due to the risk involved in testing or deploying such systems. Accessible platforms have the potential to widen the field to a broader set of researchers and students. Existing efforts in making on-road autonomy more accessible have seen success, yet aggressive off-road autonomy remains underserved. We seek to fill this gap by introducing HOUND, a 1/10th-scale, inexpensive, off-road autonomous car platform that can handle challenging outdoor terrains at high speeds. To aid development speed, we integrate HOUND with BeamNG, a state-of-the-art driving simulator to enable both software in the loop as well as hardware in the loop testing. To reduce the extent of ruggedization required, and thus cost, we integrate a rollover prevention system as a safety feature into the platform. Real-world trials over 50 kilometers demonstrate the platform's longevity and effectiveness over varied terrains and speeds. Build instructions, datasets, and code disseminated via: https://sites.google.com/view/prl-hound/home

Figures (10)

• Figure 1: HOUND has been tested on dirt hills, grasslands, gravel trails, and tarmac.
• Figure 2: Autonomy stack components, specified with (rate in Hz, latency in ms), with $\approx$ implying estimated latency based on maximum update rate. SITL spoofs perception, useful for isolating control problems, whereas HITL spoofs sensors, useful for testing deployed software.
• Figure 3: Illustration of the coordinate frame used by HOUND
• Figure 4: BeamNG's soft-body physics simulation (\ref{['fig:beamNG_sim']}) allows accurate simulation of second-order inertial effects, useful for problems such as rollover prevention(\ref{['fig:isolated rollover']}). The integration with BeamNG spoofs the elevation map(\ref{['fig:offroad_sim']}) for the off-road environment to isolate the control problem.
• Figure 5: Static limiter obtains lower $A^b_y/A^b_z$ in \ref{['fig:lat_ratio']} and more rollovers(\ref{['fig:rollover_rate']}). In 3 out of 4 scenarios, the rollover rate is 0 for both static limiter and full RPS in \ref{['fig:rollover_rate']}. \ref{['fig:rollover_rate_speed']}, \ref{['fig:rollover_vert_acc']} show the breaking of instantaneous steering and constant ground contact assumption.