Organizations: School of Mechanical Engineering, Beijing Institute of Technology, Beijing, China, 100081. · Hong Kong Polytechnic University, 11 Yuk Choi Road, Hung Hom, Kowloon, Hong Kong, 999077. · China North Vehicle Research Institute, Bejing, China, 100072.
Traversability estimation is critical for ground vehicle teleoperation, yet existing methods often neglect the human operator's preferences for speed and risk tolerance. To address this gap, we propose PATH-Net, a Preference-Aware Traversability estimation Hierarchical Network that adaptively generates preference-conditioned traversability cost maps and speed profiles for off-road teleoperation. Our framework integrates visual and geometric bird's-eye-view features with short-term historical vehicle states through a multi-head shared-backbone architecture. Our PATH-Net infers traversability cost maps while optimizing speed recommendations by using maximum entropy inverse reinforcement learning and sparse speed supervision. Our method incorporates operator preference via an asymmetric feature aggregation mechanism, splitting inference into a cost branch that prioritizes high-risk traversability regions and a speed branch that outputs preference-adjusted recommended speeds. Experiments with 15 operators across 225 teleoperated trajectories show alignment with user-specific navigation strategies, achieving modified Hausdorff distance of 0.17 m (aggressive) and 0.15 m (conservative). In the questionnaire assessment, participants reported lower NASA-TLX workload scores with PATH-Net recommendations than without recommendations, decreasing from 34.7 to 22.2. These results support preference-conditioned trajectory and speed recommendation for off-road teleoperation.
Figures & tables
Fig. 1: Our PATH-Net integrates environmental factors, kinematics, and expert preferences to provide off-road traversability recommendations.
Fig. 2: Overview of PATH-Net. Visual-geometric BEV features and recent vehicle states history are fused to predict candidate cost and speed maps, which are then converted into preference-conditioned cost and speed maps by upper-tail cost aggregation and quantile-based speed selection.
Fig. 3: The spatiotemporal feature fusion module. The geometric guidance map is fused with visual features to form the environment feature map, while the kinematic history is encoded by a TCN and spatially tiled. Their concatenation yields the fused feature map used for subsequent cost and speed prediction.
Preference Category
Assigned ρ
Aggregation Level α
Participants
Aggressive
−0.9
0.05
4
Neutral
0.0
0.50
6
Conservative
+0.9
0.95
5
Total
-
-
15
TABLE I: Questionnaire-Defined Preferences and Aggregation Parameters
Fig. 4: Experimental hardware and teleoperation system.
Fig. 5: Training convergence and validation-trajectory alignment. (a) Training objectives of PATH-Net over 50 epochs, including the total objective L , IRL loss LIRL , and speed regression loss Lspeed . (b) Validation pMHD of PATH-Net and the adapted baselines under the same 50-epoch training budget.
Fig. 6: Qualitative comparison of cost-related and speed-related outputs for a representative held-out off-road test case. The second row shows the final costmaps, the third row shows the corresponding speed-related outputs. PATH-Net and Velociraptor generate spatial speedmaps, whereas HDIF generates a costmap using a fixed speed query, and its speed-related visualization is derived from the common planner rather than a learned speedmap.
Method
Metric
Aggressive
Conservative
Velociraptor
pMHD (m)
0.37±0.08
0.36±0.07
sMHD (m/s)
0.23±0.06
0.21±0.05
HDIF
pMHD (m)
0.39±0.09
0.38±0.08
sMHD (m/s)
–
–
PATH-Net, neutral
pMHD (m)
0.31±0.06
0.29±0.06
sMHD (m/s)
0.18±0.04
0.17±0.04
TABLE II: Trajectory Alignment with Held-out Demonstrations (mean ± SD over Test Trajectories; lower is better).
Fig. 7: Dataset-level operating points comparison. Human demonstrations are shown by their executed speed and path-averaged cost. Each PATH-Net marker shows the mean recommended speed and mean path-averaged cost obtained by applying one preference setting to the same recorded cases. The curve indicates the non-dominated frontier of the human demonstrations.
TABLE III: NASA-TLX Workload Ratings under Different Recommendation Conditions. SD values are shown in gray parentheses.
Preference Setting ( ρ )
Min. Clearance (m)
Avg. Recommended Speed (m/s)
Path-Averaged Cost
Aggressive ( −0.9 )
0.18
1.07
0.36
Neutral ( 0.0 )
0.35
0.68
0.22
Conservative ( +0.9 )
0.62
0.43
0.14
TABLE IV: Recommended Trajectory Measures under Different Preference Settings. Values are averaged over held-out test cases.
Fig. 8: Examples of PATH-Net recommended trajectories under different preference settings with the historic trajectory (white curve), held-out demonstration (red curve), and PATH-Net recommended trajectory (black curve).
Fig. 9: Aggregated costmaps and speedmaps under different preference settings. All costmaps use the range [0,1] , and all speedmaps share the range [0,vmax] . Moving from the aggressive to the conservative setting increases the contribution of larger candidate costs and lowers the recommended speed bound.
Teleoperation, a human-in-the-loop control scheme, allows a human operator to remotely command and guide a mobile robot to navigate in off-road environments, yet fluent and user-friendly tele-navigation requires an alignment of traversability evaluation between human and robot. In the teleoperation system, the human operator typically utilizes off-site incomplete and delayed feedback via a human-machine interface to make a judgment of traversability, while the robot makes such an evaluation based on in situ onboard sensory information, which could cause divergent traversability estimation and thus generate mismatched decisions and actions. Existing approaches for traversability modeling, estimation, and prediction are mainly derived from the view of robots, i.e., robot-centric, and are practically suitable for fully autonomous mobile robots, but neglect the influence of human operators. To address this problem, this paper extends the concept of traversability from robot-centric to human-centric by accounting for the operator's cognitive states, such as attention, workload, and risk tolerance or awareness, termed tele-traversability. We first revisit the definitions and roles of traversability in robotics and then extend them to teleoperation settings. Finally, we highlight future trends and open challenges of tele-traversability toward human-centric teleoperation systems.
Lewei Feng, Qi Chen, Wenshuo Wang +6
School of Mechanical Engineering, Beijing Institute of Technology, China · Faculty of Marine Science and Technology, Beijing Institute of Technology, China · Department of Civil and Environmental Engineering, University of Auckland, New Zealand +2
In vision-based off-road navigation, a robot needs to know not only which obstacles to avoid but also which terrain is better. The first is handled by freespace detection or semantic segmentation. The second is usually answered with a traversability score, but no universal ground truth exists for such a score, so perception falls back on a predefined value per semantic class or a freespace confidence. These scores say what a region is, not which region a robot should prefer. We therefore formulate this preference as visual traversability ranking, an ordering of visible terrain that can be supervised by comparisons between two regions. Standard annotations do not label preference, but they imply its direction. We present TravPro, which converts these annotations into ordered region pairs and fits a small readout on frozen vision--language model (VLM) patch tokens to these pairs. The tokens are clustered once into a fixed prototype bank, and the readout learns a preference score per prototype. The readout is then applied to every patch and serves as a teacher that turns sparse comparisons into dense preference pseudo-labels without pixel-wise annotation. An RGB student distills these maps into a dense terrain-preference map together with a non-ground mask that excludes obstacles and background from the ranking. On five unseen domains, TravPro reaches a mean pairwise accuracy of 0.915 against 0.783 for the strongest baseline, producing an ordering sensitive to surface condition that a per-class value cannot represent. The same VLM and the same supervision yield no such ordering when the VLM is prompted and the supervision is used as dense targets; what matters is how they are used.
Reliable traversability estimation is essential for autonomous navigation on unstructured terrain. State-of-the-art approaches typically evaluate traversability from deterministic terrain elevation, thereby overlooking spatial correlations in the terrain and failing to account for how elevation uncertainty propagates through nonlinear traversability mappings. We propose ProTrav, a probabilistic terrain representation that uses function-space sampling to generate spatially correlated terrain realizations and propagate elevation uncertainty to traversability cost. First, a Sparse Variational Gaussian Process (SVGP) reconstructs a probabilistic elevation field from observations. Then, posterior function-space sampling generates spatially correlated terrain realizations, preserving spatial dependencies between neighboring locations while propagating elevation uncertainty through geometric features into traversability cost. Consequently, a distribution of traversability cost at arbitrary location is established, where the mean and standard deviation quantify the expected traversability cost and associated uncertainty, respectively. A risk term is defined by moment-based Upper Confidence Bound (UCB) and incorporated into a Model Predictive Path Integral (MPPI) controller for risk-aware navigation. Simulation experiments demonstrate navigation success rates exceeding 90% on different terrains, compared with below 40% for a baseline, while reducing attitude variations. Furthermore, the proposed approach is validated on a robot traversing real-world off-road terrain.
Chuanlin Zuo, Miloš Prágr, Arun Kumar Singh +2
Faculty of Engineering and Natural Sciences, Tampere University, Finland. · Institute of Technology, University of Tartu, Estonia. · AI and ML Department, Volvo Autonomous Solutions, Sweden.