Humanoid-Bimanual-GraspStability

Physics-informed (friction-cone + reactive tactile grip-force control) bimanual object-manipulation dataset: 20 end-effector embodiments, 800 trials across lift/carry/handover/rotation tasks, with slip/drop/misalignment failure-outcome labels.

About Dataset

Humanoid-Bimanual-GraspStability simulates two-handed object-manipulation trials -- lifting, carrying, handover between hands, and in-hand rotation stress tests -- across 20 independently-sampled end-effector embodiments (max grip force, fingertip friction coefficient, tactile sensing latency, grip-force control bandwidth, sensor noise). Each hand's grasp is modeled as a friction-cone contact interface: a grasp sustains tangential (shear) force up to the friction coefficient times the normal (grip) force before slipping. Grip force is regulated by a reactive, tactile-triggered controller modeled on published human-inspired robotic grasp-control research, so most trials recover from transient slip rather than failing outright. Trials additionally vary object mass, center-of-mass offset, surface-friction perturbations, sudden dynamic load-shift events, and external bump disturbances. Each timestep records 20 channels per hand-pair: grip/normal/shear force, slip velocity and cumulative slip, object roll/pitch drift, IMU-like object acceleration, wrist torque, and a continuous grasp-stability margin. Trials are labeled with one of four outcomes: success, slip_recovered, slip_dropped, or misaligned_dropped. This is a fully synthetic dataset -- no real robot telemetry, proprietary, or personal data was used.

Purpose of Dataset

This Dataset Supports Research On Bimanual Grasp-failure Prediction And Reactive Grip-force Control That Generalizes Across Different End-effector Designs, Rather Than Being Tuned To One Specific Gripper Or Hand. It Targets Researchers And Engineers Building Tactile-sensing-based Slip-detection Systems And Grasp-stability Monitors For Humanoid Or Bimanual Manipulation Platforms, Particularly Where Early Detection Of Incipient Slip (Before An Object Is Actually Dropped) Is The Operationally Useful Capability. The Embodiment-grouped Design Supports Studying Whether A Grasp-failure Predictor Trained On Some Hand Designs Transfers To A New Gripper It Has Never Handled A Single Trial From.