Humanoid robots are expected to traverse complex terrains, where the plantar support may vary dramatically due to foot placement errors, ground properties, and transient dynamics. To achieve robust locomotion, the robots are required to adapt to uneven terrain and uncertain foot--ground interactions. Existing locomotion policies rely primarily on proprioception or exteroceptive terrain perception, where the former provides only indirect evidence of plantar support, while the latter predicts contact conditions before touchdown but cannot observe the actual support in real-time. Although some studies incorporate plantar contacts as an auxiliary perception, they rely mainly on summary statistics, overlooking the spatial topology of plantar pressure, which provides a more direct characterization of the realized contact state. To bridge this gap, we present Tac4Loco, a tactile-perceptive framework that incorporates multi-array plantar pressure as direct feedback for humanoid locomotion. We formulate a topology-preserving ordinal representation to map simulated and physical sensor signals into a shared observation space, with a dual-branch encoder for extracting their spatial and temporal representations. Subsequently, the learned spatiotemporal features are integrated with augmented proprioception including terrain estimation cues, and provided to an asymmetric actor-critic architecture for policy learning. Extensive simulation and real-world experiments demonstrate improved tracking performance and support adaptation on terrains with inclined, partial, asymmetric, and changing support. We further demonstrate its zero-shot deployment on unseen compliant and unstructured terrains, including a foam platform and a gravel road. All code and experimental configurations will be released as open-source to facilitate reproducibility.
Figures & tables
Fig. 1 : Real-world deployment of Tac4Loco with plantar pressure feedback. (a) Unitree G1 equipped with bilateral plantar pressure insoles. (b) Representative comparison on a compliant terrain transition, where Tac4Loco succeeds while the proprioception-only baseline fails. (c) Selected support states and their corresponding plantar pressure maps, highlighting distinct pressure topologies under different contact conditions.
Fig. 2 : Overview of the Tac4Loco pipeline. During training, proprioception and raw plantar pressure measurements are collected in simulation. The pressure signals are preprocessed through matrixization, history extraction, terrain-orientation estimation, and observation augmentation, and then encoded by a dual-branch plantar pressure encoder consisting of a spatial branch and a temporal branch. The resulting tactile representation is fused with proprioception in an asymmetric actor–critic framework for policy learning. After training, the learned actor is transferred to the real robot and driven by physical plantar pressure measurements processed under the same representation.
Fig. 3 : Architecture of the dual-branch plantar pressure encoder, comprising the state-conditioned spatial branch (a) and temporal pressure–proprioceptive branch (b).
Fig. 4 : Representative simulation terrains and the corresponding bilateral plantar pressure maps of Tac4Loco: (1) flat terrain, (2) gently undulating terrain, (3) V-trench terrain, (4) slope terrain, and (5) random support-height terrain.
Terrain
Method
Survival (%) ↑
Lin. Vel. Err. (m/s) ↓
Yaw Vel. Err. (rad/s) ↓
FL
Proprio-only
100.0
0.337 ± 0.244
0.380 ± 0.224
Tac4Loco
100.0
0.131 ± 0.024
0.329 ± 0.189
GU
Proprio-only
100.0
0.381 ± 0.252
0.373 ± 0.235
Tac4Loco
100.0
0.140 ± 0.030
0.339 ± 0.185
RSH
Proprio-only
71.7
0.351 ± 0.248
0.416 ± 0.207
Tac4Loco
96.5
0.168 ± 0.051
0.378 ± 0.159
TABLE I : Overall simulation performance of Tac4Loco.
Terrain
Proprio-only
Tac4Loco
FL
1.434 ± 0.065
0.495 ± 0.048
GU
1.333 ± 0.520
0.551 ± 0.258
RSH
1.095 ± 0.482
0.545 ± 0.337
TABLE II : Open-terrain drift angle under translational commands.
Fig. 5 : Step length under challenging support conditions. Error bars denote standard deviation.
Method
Task Performance
Gait and Motion Metrics
GU Lin. Vel. Err. (m/s) ↓
RSH Surv. (%) ↑
SP Surv. (%) ↑
VT Surv. (%) ↑
RSH Step Interval CV (-) ↓
SP Step Len. (m)
VT Lateral Drift (m) ↓
w/o Pressure Enc.
0.158 ± 0.041
90.6
73.4
68.8
0.228 ± 0.109
0.421 ± 0.055
0.194 ± 0.077
w/o Terrain Est.
0.228 ± 0.061
94.9
42.2
96.9
0.201 ± 0.102
0.267 ± 0.063
0.067 ± 0.029
Tac4Loco
0.140 ± 0.030
96.5
77.9
100.0
0.195 ± 0.098
0.452 ± 0.047
0.059 ± 0.023
TABLE III : Ablation study of Tac4Loco components.
Fig. 6 : Real-world deployment validation of Tac4Loco. (A) Static structural terrain (covered by simulation): (1) forward traversal over a 9∘ ramp and transition from its upper platform to lower flat ground; (2) lateral ascent onto a 9∘ ramp; (3) transition from a 15∘ ascending ramp to a 9∘ descending ramp; (4) traversal of a V-trench formed by two 9∘ ramps. (B) Deformable and granular terrain (zero-shot): (1) transition from rigid flat ground onto a compliant foam platform; (2) transition from a 9∘ ramp onto the compliant platform; (3) gravel road.
Configuration
Support challenge
Completion
Proprio-only
Tac4Loco
9∘ ramp edge
Partial support / drop
7/10
10/10
Lateral 9∘ ascent
Asymmetric support
1/10
8/10
15∘ up + 9∘ down
Slope transition
0/10
10/10
9∘ V-trench
Edge support
–
10/10
Flat → foam
Compliant transition
0/10
7/10
TABLE IV : Real-world deployment comparison of Tac4Loco and the proprioception-only baseline.
Humanoid parkour policies can traverse various terrains, but task completion may mask challenges of harsh landings, edge contacts, and unstable stance contacts. Humans naturally regulate foot-terrain interaction through tactile feedback, modulating contact compliance according to terrain stiffness. This highlights a key domain gap between humans and humanoid robots: the absence of rich tactile sensing in most humanoid systems. We address this problem with TactileStep, a deployable tactile learning framework that brings sole pressure sensing into humanoid locomotion control for softer touchdowns and more stable support. TactileStep aligns tactile simulation with the real pressure insole, allowing the policy to learn from the same contact features available on hardware. During training, we use tactile and motion cues to recognize different foot-contact phases and apply phase-aware rewards that encourage safer landing and more stable stance. Evaluated in simulation and on a Unitree G1 humanoid across diverse terrains, TactileStep reduces peak touchdown force by up to 48.8% and peak A-weighted impact noise by up to 30.1 dB over a strong perceptive baseline, while increasing stance contact area by up to 23.8%.
Learning-based control has revolutionized dynamic locomotion, yet navigating unstructured terrain remains limited by a robot's incomplete awareness of imminent ground contact. While global perception systems such as LiDARs and depth cameras provide environmental context, they are frequently plagued by latencies, occlusions, and the high computational cost of dense geometric reconstruction. On the other hand, proprioceptive feedback is purely reactive, initiating corrections only after impact has occurred. This work explores embedding a minimal suite of low-cost, high-frequency infrared proximity sensors directly into the feet of a quadrupedal robot. These sensors provide "pre-contact" feedback that is robust to self-occlusions and significantly less computationally demanding than conventional vision-based pipelines. By integrating these localized signals into a reinforcement learning framework, we enable the robot to anticipate terrain discontinuities such as gaps and stepping stones that are problematic for traditional perception stacks due to occlusions or state estimation drift. We demonstrate that such sparse, near-field sensing can be reliably modeled in simulation and transferred to the real world with high fidelity. Experimental results show that local proximity sensing substantially improves traversal robustness over discrete terrain and offers a low-power, low-latency alternative or complement to complex global perception suites in unpredictable environments. For more information about results and methods, please see the project website: https://sites.google.com/view/foot-tof/home.
Jiale Fan, Connor Flynn, Tianao Xu +4
All authors are with the Robotic Systems Lab, ETH Zurich.
Humanoid locomotion over complex terrain requires anticipating footholds that may no longer be visible at touchdown. Limited camera coverage and self-occlusion make it necessary to retrieve relevant terrain information from earlier observations. We present FootQuery, a perceptive locomotion framework that queries depth history using each foot's predicted next touchdown. The policy predicts touchdown locations and uncertainty from proprioception and uses these distributions, together with per-foot features, to query sparsely sampled historical depth frames. During training, realized contacts are projected into historical images to supervise retrieval at the regions where those contacts were visible. The retrieved per-foot features are fused with global visual memory to generate control actions. A progressive force-assistance curriculum supports early exploration, while event-consistent tread-midline shaping encourages coordinated stair contacts. Deployment requires only proprioception and onboard depth images. In simulation, the complete framework outperforms its component ablations on the most challenging tested stairs, gaps, and platforms. Real-world experiments on a Unitree G1 demonstrate continuous traversal with a single policy across outdoor stairs and indoor routes combining stair ascent and descent, platforms, and gaps. These results support organizing visual history around anticipated contacts for perceptive humanoid locomotion.
Tao Dong, Jia Yu, Yuxuan Fan +5
Institute for AI Industry Research (AIR), Tsinghua University, Beijing, China. · University of Science and Technology Beijing, Beijing, China. · Nanyang Technological University, Singapore. +1