FutureRay: Control-Aligned Future Range for Agile Quadruped Navigation

Tianhao Zang; Shanze Wang; Ziqian Wang; Liyou Luo; Zihan Liu; Xingjian Xie; Wei Zhang

cs.ROSep 26, 2026

FutureRay: Control-Aligned Future Range for Agile Quadruped Navigation

Authors: Tianhao Zang, Shanze Wang, Ziqian Wang, Liyou Luo, Zihan Liu, Xingjian Xie, Wei Zhang

Organizations: College of Information Science and Technology, Eastern Institute of Technology, Ningbo, P. R. China. · University of Maryland, USA. · The Hong Kong Polytechnic University, Hong Kong. · School of Computer Science, University of Nottingham, UK. ·

Abstract

Moving obstacles can block a previously clear route while a quadruped robot executes a motion command. We investigate whether predicting changing clearance improves navigation when motion selection accounts for the robot footprint and the time needed to react and brake. We present FutureRay, which predicts ranges across viewing directions and future times, together with encounter risk, from depth-derived range history and observable robot motion. Training emphasizes near-term clearance and penalizes errors that overstate available space. A local planner queries the same forecast for candidate headings and combines it with current observations to check clearance around the robot footprint. Model-based reaction--braking limits guide speed selection, and the resulting velocity commands are passed to a fixed locomotion policy. In paired evaluations on 60 static and dynamic simulation scenes, FutureRay achieves 93.3% completion, compared with 75.0% for current-range persistence and 80.0% for Cartesian Kalman rollout, with perception, planning, and locomotion held fixed. FutureRay also records fewer collisions than both baselines. Qualitative trials on a physical quadruped show avoidance initiated while an obstacle is approaching the route, followed by renewed goal progress. These results show that joint range and encounter-risk prediction can improve obstacle avoidance without retraining the locomotion policy.

Figures & tables

Fig. 1: Future sensor space for anticipatory avoidance. The schematics contrast a response to current clearance with a detour initiated from a forecast of an approaching crossing. The physical quadruped robot sequence illustrates onboard avoidance and subsequent progress toward the goal.

Fig. 2: FutureRay training and control. Depth-derived range history and observable ego motion produce a future-range field and encounter risk. Current geometry, physical constraints, and temporal utility determine candidate body velocities. Command-history regularization and rate limits precede the fixed quadruped locomotion policy. Supervision is confined to the offline training path. Here, $r_{t}^{90}=R_{t}$ , and $\widehat{R}_{t}$ denotes the future-range field issued at time $t$ .

Fig. 3: Forecast-to-control diagnostic from one high-dynamic quadruped rollout. (a) Current-range persistence, the learned nominal forecast, and offline later-pose geometry at

t=2.02

s share bearing–time axes. The center range is 5.27 m; its

+0.50

-s prediction and later geometry are 0.99 and 0.91 m. (b) The logged heading bank marks early corridor proximity with crosses and selects

+35^{\circ}

at 1.11 m/s. (c) Synchronized center-range and issued body-command traces show the encounter; issued commands include component and rate limits. Later geometry is evaluation-only and uses the executed future robot-root positions and yaw angles.

Method	S $\uparrow$	C $\downarrow$	$d\uparrow$	$v_{x}\uparrow$ (m/s)
Reactive persistence	75.0	20.0	0.333	1.321
Kalman rollout	80.0	11.7	0.455	1.121
FutureRay (ours)	93.3	1.7	0.534	1.164

TABLE I: Shared-planner quadruped navigation comparison (

n=60

). S/C: completion/collision (%);

d

: mean minimum clearance (m);

v_{x}

: body-forward speed after 2 s (whole episode if shorter). All tables shade FutureRay and bold metric bests, including ties.

Method	S $\uparrow$	C $\downarrow$	T $\downarrow$	$v_{x}\uparrow$	$\mathrm{S}_{\rm HD}\uparrow$
SRU [ 26 ]	56.7	30.0	13.3	1.207	20.0
CE-Nav [ 27 ]	66.7	33.3	3.3	1.235	80.0
REASAN [ 6 ]	85.0	13.3	1.7	1.231	80.0
NavRL (planar) [ 28 ]	70.0	5.0	25.0	0.793	70.0
CrowdSurfer [ 29 ]	0.0	3.3	98.3	0.632	0.0
SanD-Planner [ 30 ]	73.3	11.7	15.0	0.847	65.0

TABLE II: Released-checkpoint transfer to the quadruped robot on the same 60 scenes. S/C/T: completion/collision/timeout (%);

v_{x}

: body-forward speed (m/s), defined as in Table I ;

\mathrm{S}_{\rm HD}

: completion (%) on 20 high-dynamic scenes.

Fig. 4: Eight selected Isaac Sim comparisons: (a)–(e) dynamic crossings and (f)–(h) static bypass. Each panel overlays eight synchronized exposures, fading from early to late, and complete executed trajectories of FutureRay (teal), reactive persistence (orange), Kalman rollout (yellow), CE-Nav (purple), and REASAN (blue). Colored terminal markers and the values below each panel report outcomes, elapsed time, and minimum clearance; independent runs share the same obstacle motion and fixed quadruped locomotion policy.

Fig. 5: Unified navigation comparison. (a) FutureRay, two shared-planner baselines, and six transferred methods on the same 60 paired scenes. The diamond denotes FutureRay; triangles and squares denote shared-planner baselines and transferred methods. Mean path speed equally averages episode distance/duration, including failures, and differs from the body-forward speed in the tables. Dashed segments mark the observed Pareto frontier, maximizing both axes. (b) Completion counts for the primary interface and released-method comparisons, with 20 scenes in each stratum. Color encodes completion percentage on a common scale.

Robot Navigation

Robot navigation studies how robots perceive, localize, plan, and move safely toward goals in static or changing environments.

See all 536 papers

SCAN-Planner: Spatial Collision-Aware Local Planning for Route-Guided Long-Range Quadruped Navigation
Jun 17, 2026Han Zheng, Zhe Chen, Yiwen Fu +2Collision Avoidance Robot Navigation
Shanghai Jiao Tong University, Shanghai, China.
Predict Before You Step: Auditable Occupancy Forecasting for Dynamic Obstacle Avoidance under Sparse Guidance
Sep 22, 2026Yuhui Mao, Fen Liu, Shenghai Yuan +3Robot Navigation 3D Occupancy Prediction
School of Electrical and Electronic Engineering, Nanyang Technological University, 50 Nanyang Avenue, Singapore 639798
Learning Agile Navigation in Crowded Environments for Quadruped Robots
Jul 16, 2026Shuyu Wu, Zeyu Liu, Tianbao Zhang +6Robot Policy Learning Robot Navigation
Shanghai Jiao Tong University, Shanghai 200240, China · State Key Laboratory of High-end Heavy-load Robots, Midea Group, Foshan 528300, China

Jun 17, 2026J/K move · Enter open · S save

SCAN-Planner: Spatial Collision-Aware Local Planning for Route-Guided Long-Range Quadruped Navigation

Han Zheng, Zhe Chen, Yiwen Fu, Ming Yang+1

Shanghai Jiao Tong University, Shanghai, China.

Collision Avoidance Robot Navigation Safe Robot Navigation Quadruped Robot Navigation Robot Motion Planning

Open paper AskPDF arXiv

Quadruped robots are increasingly expected to navigate through narrow passages, cluttered indoor scenes, and large-scale 3D unstructured environments. Existing local planners commonly approximate the robot using isotropic geometric inflation or rely on planar and elevation-map representations, leading to conservative motion in tight spaces and limited reasoning about overhanging structures. This letter presents SCAN-Planner, a spatial collision-aware local planning framework for long-range quadruped navigation. A yaw-aware twin-cylinder footprint is used to model the elongated robot body, enabling whole-body collision evaluation through sparse queries in an inflated 3D occupancy map. We further introduce a projected A* search that generates collision-free guidance on an interpolated ground-following surface, with z-gradient suppression to avoid obstacles horizontally while maintaining vertical stability. For large-scale deployment, a robot-centric sliding map with boundary fallback provides high-resolution local collision checking and recovery from local dead ends. Simulation and real-world experiments demonstrate that SCAN-Planner generates safe, smooth, and efficient trajectories in dense clutter, 3D unstructured scenes, stair traversal, and long-range navigation tasks.

Abstract

Robot Navigation

Explore similar work

SCAN-Planner: Spatial Collision-Aware Local Planning for Route-Guided Long-Range Quadruped Navigation

Predict Before You Step: Auditable Occupancy Forecasting for Dynamic Obstacle Avoidance under Sparse Guidance

Learning Agile Navigation in Crowded Environments for Quadruped Robots