A fallen humanoid beneath an overhead obstacle faces a coupled recovery problem. It must create clearance while organizing support for a stable rise, since attempting to stand before making room can further increase obstruction. We introduce a reinforcement learning method with contact-supported geometric escape guidance, which rewards incremental reduction of robot-obstacle overlap without prescribing an escape path. To achieve this, a learned motion prior, together with posture-dependent support and joint-stall costs, is proposed to shape a single policy that learns when to clear the obstacle, how to establish support, and when to complete the rise, all using proprioception alone. In simulation, it recovers in 88.5% of trials beneath plates that can only be lifted, compared with 18.9% without geometric guidance, and in 98.6% of trials beneath free plates with randomized size, mass and placement. On a Unitree G1, the deployed policy recovered in 95 of 150 attempts across seven real-world obstacle scenes, including furniture and multi-object arrangements absent from training. Project page: https://lu-yidan.github.io/ROOF/
Figures & tables
Fig. 2: Training overview. Top: example initial states from each L/M/H group and fallen-posture family. Bottom: reward groups active in each training stage, in the order and colors of Table IV-B ; dashes mark absent groups. The actor and its observation normalization carry over between stages, and a single actor is deployed.
Stage
Scene share
L
M
H
Stage I (Basic)
Flat: 100%
20
40
40
Stage II (Refine)
Flat: 100%
40
40
20
Stage III (Adapt)
Flat: 50%
40
40
20
Guided plate: 25%
100
0
0
Free plate: 25%
100
0
0
TABLE I: Initial-state distribution: low ( L ), intermediate ( M ), and near-standing ( H ) percentages within each scene type.
Term
Definition
I: Basic
II: Refine
III: Adapt
TABLE II: Reward definitions and signed weights. Colors match Fig. 2 ; dashes indicate inactive terms. Weights are multiplied by the 0.02 s control period. Notation. [x]+=max(x,0) , χ(x)=clip(x,0,1) , ℓ(x;a,b)=χ((x−a)/(b−a)) , σ(x)=x2(3−2x) . t : control step; s : recovery phase; j : joint; on/full : ramp onset/saturation; utorso=χ(−gzb) ; vfoot2=meani∈feet∥vi∥2 . target , up/down , and limit denote posture targets, task speed bounds, and execution speed bounds; δh : taper margin; eknee : knee-range violation; ks : phase scales. Δat=at−at−1 and Δ2at=Δat−Δat−1 . Loading. ρj=∣τj∣/τjmax ; Tj : time above ρon ; q˙stall : stall-speed threshold; Aj : recent joint excursion; mjload : low-pass filtered ρj2 . Jhigh selects the three largest effort costs; ⟨⋅⟩ averages physics substeps. Escape gates. Binary indicators select episodes containing a plate ( δplate ), plate states passing penetration, contact-force and initial-contact checks ( δtvalid ), prior plate contact in the episode ( δtcontact ), and currently confirmed escape ( δtescape ). gtprogress=δtvalidδtcontact(1−δtescape) ; fthand is the fraction of hands providing support. ∗ : terms summed in Rgeom ; † : scaled by min(nupdate/500,1) in Stage II, fully active in Stage III.
School of Control Science and Engineering, Shandong University, Jinan, 250061, China · Key Laboratory of Machine Intelligence and System Control, Ministry of Education, China