Tasks

Fold Laundry

Pick up a rumpled garment from a flat surface and fold it into a tidy rectangle. A canonical dexterous manipulation benchmark — deformable object, long horizon, multi-step.

Pick and Place Cube

Grasp a small cube from a known location and place it at a target pose on a flat tabletop. Canonical short-horizon manipulation benchmark.

Bimanual Peg Insertion

Two arms coordinate to pick up a peg and insert it into a socket held by the other arm. Tests fine-grained force control and cross-arm coordination.

Walk on Flat Terrain

Traverse a flat, unobstructed surface from start to goal. Baseline locomotion capability every legged platform should pass.

Climb Stairs

Ascend or descend a standard residential staircase. Requires terrain perception and adjusted gait to handle discrete steps.

Recover from External Push

Remain upright (or regain footing) after a sudden lateral perturbation. A standard robustness benchmark for legged control policies.

Warehouse Aisle Navigation

Drive through a narrow warehouse aisle avoiding shelves, forklifts, and people. Tests SLAM, dynamic obstacle avoidance, and long-horizon route planning.

Drone Field Survey

Fly a gridded pattern over an outdoor field and capture overlapping RGB imagery for downstream photogrammetry. Multi-leg waypoint following with wind disturbance.

Drone Follow a Person

Track a walking person at fixed offset and altitude through a cluttered outdoor environment. Tests vision tracking and reactive control under occlusion.

Solve a Rubik's Cube

Single-handed or bimanual manipulation of a 3x3 Rubik's Cube from scrambled to solved. Open problem: long-horizon dexterity, contact-rich reorientation.