Legged robots learn to avoid fragile surfaces during walking
SABER: Learning Attention-based Semantic Affordance for Legged Locomotion
Robotics
Summary
Robots that walk on legs often rely on maps of the ground’s shape to choose safe steps. However, some surfaces might look walkable but can break or be unsafe to step on. The authors built a robot control method called SABER that not only looks at the shape of the ground but also understands what kind of surface it is, like fragile boxes or pipes. SABER helps the robot pick steps that avoid these unsafe surfaces while moving stably. They tested this method on a real robot, showing it could walk safely over different kinds of obstacles indoors and outdoors.
What this means in practice
- •For industrial robotics teams: Enable legged robots to avoid stepping on fragile equipment by integrating terrain meaning into walking policies.
- •For outdoor robot operators: Improve safe navigation of legged robots over varied terrain by identifying and avoiding semantically hazardous footholds.
Authors
Hari Prasanth Palanivelu, Samuel Sze, Kennard Garrison Johannes, Albertus Hendrawan Adiwahono, Meng Yee, Chuah
Abstract
Perceptive legged locomotion has advanced rapidly by integrating terrain geometry into learned policies, yet the integration of terrain meaning remains sparse: a pipe, a patch of grass, or a fragile box may be geometrically traversable while being inappropriate for contact. In industrial environments, where legged robots increasingly operate, a single misplaced step can damage fragile equipment, destabilize the robot, or endanger the site. To address this, we introduce SABER, a planner-free reinforcement-learning policy that jointly reasons about terrain geometry and semantic contact permission. The policy consumes a unified terrain-affordance map, where each cell encodes local 3D geometry and a semantic contact cost. We augment cross-attention with a learned, signed semantic bias: an additive term on the attention logits, gated by the contact cost, that reweights flagged cells by their distance from the nearest foot. A hazard therefore reshapes attention where it can still affect the next foothold, and its influence fades where it cannot. The resulting policy selects footholds on permitted support and keeps the leg clear of forbidden regions throughout the swing phase. We perform a systematic ablation that isolates the contribution of each architectural component; removing the semantic bias alone increases forbidden contacts by 55% while velocity tracking is unchanged. We validate the policy on a Unitree B2, demonstrating sim-to-real semantic contact selection across indoor and outdoor environments and four semantic obstacle classes.