Organizations: Department of Electrical and Computer Engineering National University of Singapore, Singapore · Department of Mechanical Engineering National University of Singapore, Singapore
Safety-critical high-speed legged locomotion requires motion planners to balance safety foresight with online computational efficiency. Although short prediction horizons reduce rollout costs, they provide limited safety foresight, and strict safety constraints severely restrict the feasible trajectory space. We propose a recovery-guided model-based diffusion (MBD) framework for safety-critical legged locomotion. It incorporates a learned terminal recovery critic into short-horizon trajectory optimization. Specifically, the critic approximates the worst-case future clearance margin along rollouts of a designated recovery policy. We introduce a feasibility-guided refinement strategy that combines within-horizon constraints and terminal recoverability at each sampling stage, while retaining feasible elites to mitigate sampling-support loss. Across 50 paired randomized Unitree Go2 obstacle-avoidance trials, our approach achieves zero collisions and 98% strict-clearance success, while requiring substantially less computation than long-horizon MBD baselines. A project page is available at https://zack4417.github.io/rg-mbd/.
Figures & tables
Figure 1: Finite-horizon safety limitation with H=10 . The yellow line shows the base trajectory. Although initially safe at 1.6m/s , the robot’s terminal state fails the safety recovery criterion.
Component
Parameter and value
Online planner
H=10,Δt=0.02sNs=2048,M=2,Minit=6
DIAL sampling
λ=0.05,ρσ=0.5σctl=[1,1,1]
Safety guidance
dsafe=0.02mE=32,Ts=0.1
Recovery policy
Hrec=50,Nrec=2048Mrec=4,Lrec=60 steps
Critic
wfs=5,κ=0.05
Critic loss
λTD=λMC=1λenv=0.25,λdec=0.10
Table 1: Key implementation parameters.
Figure 2: Unitree Go2 obstacle-avoidance snapshots: (a)–(c) DIAL with H=10 , 20 , and 50 ; (d) recovery-guided MBD with H=10 . Yellow curves show the executed base trajectory, translucent poses show closed-loop states at 20-step intervals, and annotations give selected longitudinal base velocities.
Method
H
rcol
dmin
rsafe
ρfeas
tcom
tp95
(%) ↓
(cm) ↑
(%) ↑
(%) ↑
(ms) ↓
(ms) ↓
DIAL
10
98.0
−2.77±2.50
0.0
–
31.0±0.1
31.2
DIAL
20
68.0
−0.13±4.07
20.0
–
113.9±0.5
114.7
DIAL
50
58.0
−0.37±3.52
22.0
–
273.3±1.5
276.1
Shield-DIAL
10
96.0
−1.78±1.74
0.0
–
31.3±0.1
31.4
Shield-DIAL
20
68.0
−0.60±3.23
20.0
–
114.2±0.3
114.4
Table 2: Quantitative comparison over 50 paired randomized obstacle-avoidance scenarios. Shield-DIAL augments DIAL with a discrete CBF post-filter on u0 ( Yin et al., 2024 ) . Dual-shield and Soft-guided are ablations. Bold denotes the best result.
Figure 3: Safety comparison of DIAL ( H=10,20,50 ) and recovery-guided MBD ( H=10 ). (a) Minimum whole-body clearance; dotted and dashed lines mark the 2cm safety margin and 0cm contact boundary. (b) Executed safety state.
Figure 4: Computational-budget ablation. (a) Mean latency; black ticks show 95th percentiles and shading marks the 20ms ( 50Hz ) real-time deadline. (b) Planning frequency versus feasible-candidate support; N∅ counts empty-support steps and Rˉ is the mean reward.
Metric
Ours
Ours + stepwise
Collision-free episodes
100%
94%
Task success
94%
66%
Feasible candidate fraction
71.72%
31.22%
Empty-support steps ( N∅ )
4
493
Mean min. clearance
2.49 cm
21.10 cm
Mean reward
−0.0586
−1.0248
Table 3: Paired ablation of recoverability filtering ( Ours ) versus the same critic with added stepwise constraints across 50 paired runs under fixed obstacle geometry.
Safe motion planning develops methods for robots and vehicles to reach goals while avoiding collisions and respecting dynamic and uncertainty constraints.
Many diffusion-based planners enforce safety through inference-time guidance, but such interleaved trajectory deformations often degrade kinematic feasibility due to manifold rupture. We propose Graph-Guided Safe Diffuser (G2SD), a hierarchical framework that leverages a high-level topological graph planner to guide a low-level diffusion model. G2SD enforces safety at a structural level by abstracting the data manifold into a learned latent graph, on which high-level planning is performed. Continuous trajectories are generated by diffusion planners, which are conditioned on the graph node representations selected by the high-level planner. Theoretical analyses demonstrate conditions under which manifold rupture occurs in diffusion planners, and show that G2SD improves safety by reducing the constraint violation probability as the number of segments increases. Experiments demonstrate that G2SD substantially outperforms baselines, increasing goal-reaching rate without any collision from 40-50% to 98% in Maze2D navigation and also achieving superior task scores in locomotion.
Nakgyu Yang, KwangBin Lee, SooJean Han
School of Electrical Engineering, Korea Advanced Institute of Science and Technology (KAIST) Daejeon, Republic of Korea
Rare but consequential failures can persist in learned locomotion policies for legged robots even when average task performance is high, in part because standard curricula primarily adapt task difficulty rather than the distribution of safety-critical experience. We introduce Predictive Safety Curricula (PSC), a framework for allocating locomotion training experience using learned predictions of future safety cost. PSC trains a distributional safety critic from policy rollouts and uses its predictions to prioritize both terrain contexts and previously encountered randomized events. The resulting curriculum modifies the training distribution while leaving the task reward and policy-optimization loss unchanged. We evaluate PSC in controlled rough-terrain locomotion and in production locomotion systems. PSC improves reliability relative to standard terrain progression, advantage-based replay, and learning-progress curricula, with the largest gains on difficult terrain and under degraded observations. The same allocation principle transfers to two production locomotion stacks. On ANYmal-D hardware, PSC reduces shank-collision incidence by 63% relative to the learning-progress curriculum across three matched training seeds, with a reduction in every seed. On a production stair-climbing platform, PSC eliminates observed shank collisions in the evaluated hardware trials. These results show that learned predictions of future safety cost can provide an effective signal for allocating training experience toward rare failure modes and improving locomotion reliability.
Ivan Ovinnikov, Pascal Sutter, Christian Gehring +1
Diffusion models have emerged as powerful tools for planning and control by learning multimodal distributions over actions and trajectories. Yet reliable inference-time safety enforcement remains a key barrier to their deployment in safety-critical tasks. Existing approaches typically project each denoising iterate onto the feasible set, even though constraints are defined only on the final clean trajectory. Enforcing feasibility on noisy intermediate samples can therefore overconstrain the sampling dynamics, substantially degrading sample quality. To address this limitation, we introduce DiRecT (Diffusion-based planning via Receding-horizon denoising with Terminal constraints), a training-free algorithm for constrained sampling from diffusion models via stochastic optimal control (SOC). DiRecT enforces constraints only on the final clean sample, avoiding unnecessary restrictions on the intermediate denoising dynamics. Inspired by model predictive control, we derive a principled receding-horizon surrogate for the otherwise intractable constrained SOC formulation, yielding an efficient algorithm that cleanly separates stochastic denoising from constraint satisfaction, progressively steering samples toward feasible final trajectories without distorting the learned diffusion dynamics. Furthermore, DiRecT is highly flexible: it can leverage off-the-shelf or domain-specific optimizers, incorporate priors over environment dynamics, and optimize additional soft rewards. Extensive experiments on safe planning benchmarks demonstrate that DiRecT substantially improves deployment safety and task performance over existing diffusion-based planning baselines.