an idea i had for an invention: the way humans can throw objects between hands, the mechanism can be leveraged for industrial and home robotics. i will provide a summary of the invention, compiled with the help of Claude Opus 4.6, with specific input from me, therefore i claim copyright (“Ashtar Ventura”, owner and webmaster of the website www.笑.wtf registered as an IDN):
SYSTEM AND METHOD FOR BIOMIMETIC DYNAMIC OBJECT TRANSFER BETWEEN ROBOTIC MANIPULATORS VIA ADAPTIVE BALLISTIC HANDOVER
CROSS-REFERENCE TO RELATED APPLICATIONS
This application claims the benefit of the filing date and is an original filing.
FIELD OF THE INVENTION
The present invention relates generally to robotic manipulation systems, and more specifically to a system and method for high-speed dynamic object transfer between two or more robotic manipulators using adaptive ballistic handover with real-time weight estimation, trajectory computation, and predictive catch synchronisation.
BACKGROUND OF THE INVENTION
In current industrial manufacturing and home robotics environments, the transfer of objects between robotic manipulators (arms, hands, grippers) is predominantly accomplished through static or quasi-static handover. In a typical static handover, a first manipulator moves an object to a pre-determined rendezvous point, holds the object stationary, a second manipulator grasps the object, and only then does the first manipulator release. This process is inherently slow and introduces significant dead time into manufacturing workflows.
Existing approaches to dynamic object transfer, such as those described in US10144591B2 (Amazon Technologies), address robotic tossing of items to fixed receiving locations (e.g., bins or shelves) within inventory systems. However, these systems do not employ a receiving manipulator that actively catches the object. They rely on passive receptacles rather than dynamic, closed-loop interception.
Academic research (e.g., “Dynamic Handover: Throw and Catch with Bimanual Hands,” arXiv:2309.05655, 2023) has demonstrated reinforcement-learning-based throw-and-catch between bimanual robotic hands. However, such systems lack pre-throw haptic mass estimation, do not perform adaptive velocity computation based on sensed object properties, and have not been integrated into industrial manufacturing pipelines.
There remains a need for a unified system that replicates the human biomechanical strategy of: (1) estimating an object’s mass through haptic interaction before throwing, (2) computing an optimal throw trajectory and release velocity based on that estimation, and (3) executing a predictive, timed catch at the receiving manipulator — all within a closed-loop control architecture suitable for production environments.
SUMMARY OF THE INVENTION
The present invention provides a system and method for transferring objects between two or more robotic manipulators via adaptive ballistic handover. The system comprises:
(a) A throwing manipulator equipped with force/torque sensors and/or tactile sensor arrays capable of estimating the mass and inertial properties of a grasped object prior to release;
(b) A ballistic trajectory computation module that receives mass estimation data, target coordinates, environmental parameters (e.g., gravitational constant, air resistance model), and geometric constraints of the receiving manipulator, and computes an optimal release velocity vector (magnitude and direction), release timing, and release point;
(c) A receiving manipulator equipped with high-speed vision sensors (e.g., stereo cameras, event-based cameras, LiDAR) and/or predictive state estimators that track the object in flight and compute the predicted intercept point and arrival time;
(d) A catch synchronisation controller that commands the receiving manipulator to pre-position and execute a timed grasp at the predicted intercept point, with real-time correction based on in-flight trajectory updates;
(e) A closed-loop feedback system that records the outcome of each transfer (success, miss, damage) and updates the ballistic model parameters, mass estimation calibration, and catch timing model via machine learning or statistical regression, enabling continuous improvement over successive transfers.
DETAILED DESCRIPTION OF THE INVENTION
1. System Architecture Overview
The system comprises at least two robotic manipulators (hereinafter “Manipulator A” and “Manipulator B”), a central or distributed compute unit, a sensor suite, and a communications bus connecting all components with deterministic, low-latency messaging (e.g., EtherCAT, TSN Ethernet, or equivalent real-time protocol).
Manipulator A and Manipulator B may be:
- Two arms of a single dual-arm robot;
- Two separate robotic arms mounted on a shared frame or production line;
- Two independent mobile robotic platforms; or
- Any combination thereof.
2. Pre-Throw Mass and Inertial Estimation (Haptic Sensing Phase)
Before initiating a throw, Manipulator A performs a haptic interrogation sequence on the grasped object. This sequence comprises one or more of the following operations:
(a) Static weighing: Manipulator A holds the object stationary and reads the force/torque sensor at the wrist or fingertip to determine gravitational force, from which mass is derived.
(b) Dynamic probing: Manipulator A executes small, controlled oscillatory or impulsive motions (e.g., a brief vertical acceleration–deceleration cycle) and measures the resulting force/torque response. The ratio of applied force to measured acceleration yields an estimate of the object’s mass and, with multi-axis probing, its rotational inertia tensor.
(c) Tactile surface characterisation: If Manipulator A is equipped with a tactile sensor array (e.g., capacitive, piezoresistive, or optical tactile skin), the system estimates surface friction coefficient and contact geometry to inform grip force planning for the throw release and for the receiving manipulator’s catch grip.
The mass estimation module outputs a tuple: (m̂, Î, μ̂, σm, σ_I, σμ) where m̂ is estimated mass, Î is estimated inertia tensor, μ̂ is estimated surface friction, and σ values are the associated uncertainty bounds.
3. Ballistic Trajectory Computation
Given the estimated object properties, the trajectory computation module solves for the optimal release state vector [x_r, v_r, t_r] (release position, release velocity, release time) such that the object follows a ballistic trajectory from Manipulator A to the intercept envelope of Manipulator B.
The computation accounts for:
(a) Gravitational acceleration (g): Standard or locally calibrated.
(b) Aerodynamic drag (optional): For lightweight or high-surface-area objects, a drag model (e.g., quadratic drag: F_d = ½ρC_dAv²) may be incorporated. The drag coefficient C_d and cross-sectional area A may be estimated from the object’s known geometry or from tactile/vision-based shape reconstruction.
(c) Manipulator B’s reachable workspace and kinematic constraints: The trajectory must deliver the object to a point within Manipulator B’s dexterous workspace, at a velocity that Manipulator B can match (i.e., the object’s arrival velocity must not exceed Manipulator B’s maximum end-effector velocity along the approach axis).
(d) Uncertainty propagation: The uncertainty bounds from the mass estimation phase are propagated through the ballistic model to produce a predicted landing distribution (e.g., a 3D Gaussian ellipsoid at the intercept plane). Manipulator B’s catch strategy is planned to cover this distribution.
(e) Safety envelope: The trajectory must not intersect any obstacle, human workspace, or exclusion zone. A collision-check module validates the computed trajectory against a real-time occupancy map before authorising the throw.
The trajectory computation may employ analytical closed-form solutions (for simple parabolic trajectories in free space) or numerical optimisation (for constrained environments with drag, spin, or obstacle avoidance).
4. Throw Execution
Manipulator A executes a planned motion profile that accelerates the object along the computed release velocity vector. At the computed release point and time, Manipulator A opens its gripper or releases its grasp in a controlled manner.
Release strategies include:
(a) Gripper opening release: A parallel-jaw or multi-finger gripper opens rapidly, releasing the object with the end-effector’s instantaneous velocity.
(b) Wrist-flick augmentation: For higher release velocities, the manipulator may execute a wrist rotation at the moment of release to add rotational velocity and stabilise the object’s flight (analogous to a human wrist flick in throwing).
(c) Finger-roll release: For multi-finger hands, a sequential finger extension imparts a controlled spin to the object, improving aerodynamic stability.
The release controller monitors the actual end-effector velocity at the moment of release and communicates the actual release state vector to Manipulator B’s tracking system, providing an initial condition for in-flight tracking.
5. In-Flight Object Tracking
Upon release, the system transitions to the tracking phase. One or more high-speed sensors observe the object in flight:
(a) Stereo camera pair or structured-light sensor providing 3D position estimates at high frame rates (≥120 Hz, preferably ≥500 Hz for short-range fast transfers).
(b) Event-based (neuromorphic) cameras providing microsecond-latency detection of moving edges, suitable for very fast objects.
(c) Time-of-flight or LiDAR sensors providing depth measurements.
The tracking module fuses sensor data with the ballistic dynamics model using a state estimator (e.g., Extended Kalman Filter, Unscented Kalman Filter, or particle filter) to produce a continuously updated prediction of the object’s intercept point and arrival time at Manipulator B.
6. Catch Synchronisation and Execution
Manipulator B’s catch controller receives the real-time trajectory prediction and commands the manipulator to:
(a) Pre-position its end-effector at the predicted intercept point, with the gripper or hand open in a configuration geometrically compatible with the incoming object’s orientation.
(b) Velocity-match its end-effector to approximate the object’s predicted velocity at the intercept point, reducing the relative impact velocity and the risk of object damage or bounce.
(c) Execute a timed grasp closure synchronised to the object’s arrival. The grasp timing is computed from the predicted time-to-intercept, the gripper’s closing dynamics (closing time, closing force profile), and a configurable safety margin.
(d) Apply adaptive grip force based on the communicated mass estimate and surface friction estimate, ensuring secure retention without crushing the object.
If the real-time trajectory prediction deviates beyond a configurable threshold from the pre-positioned intercept, Manipulator B executes a corrective motion to adjust its intercept point. If the deviation exceeds the correctable range, the system triggers a miss-abort protocol (see Section 8).
7. Closed-Loop Learning and Calibration
After each transfer, the system records:
- Actual vs. predicted intercept point and time
- Catch success or failure
- Object damage assessment (if sensors are available)
- Actual object mass (measured post-catch by Manipulator B’s force sensors)
This data is fed into a transfer model updater that refines:
(a) Mass estimation calibration parameters (correcting systematic biases)
(b) Ballistic model parameters (drag coefficients, release timing offsets)
(c) Catch timing model (gripper closing delay, sensor-to-actuator latency)
(d) Object-specific profiles (for known recurring objects in a production line)
The learning system may employ parametric regression, Bayesian updating, or neural network–based model refinement. Over successive transfers, the system converges to higher success rates, tighter intercept distributions, and faster cycle times.
8. Safety Systems
The invention incorporates the following safety provisions:
(a) Pre-throw safety check: Before every throw, the system verifies that the computed trajectory does not enter any human-occupied zone (detected via presence sensors, light curtains, or safety-rated vision). If a human is detected in the trajectory path, the throw is aborted and the system falls back to static handover.
(b) Miss-abort protocol: If in-flight tracking predicts that the object will miss the catch envelope, Manipulator B retracts and the system activates a passive catch mechanism (e.g., a safety net, padded tray, or retaining wall) to arrest the object without damage or hazard.
(c) Energy limiting: The maximum throw velocity is constrained such that the kinetic energy of any thrown object does not exceed a configurable threshold (e.g., consistent with ISO/TS 15066 collaborative robot power and force limiting guidelines).
(d) Object suitability filter: The system maintains a classification of objects that are approved for ballistic transfer. Objects that are fragile, hazardous (e.g., sharp, chemical), or outside the validated mass/size range are excluded and transferred via conventional static handover.
9. Industrial Application Examples
(a) Assembly line: On a production line, a dual-arm robot uses the invention to transfer components from a parts tray (accessed by the left arm) to an assembly fixture (accessed by the right arm). The ballistic transfer saves 0.5–2.0 seconds per cycle compared to static handover, improving throughput by 10–30% depending on the transfer distance and object mass.
(b) Warehouse order fulfilment: Two robotic arms stationed at adjacent packing stations toss lightweight items (e.g., boxed consumer goods) between them to balance workload dynamically, without requiring conveyor interconnection.
(c) Home robotics: A domestic robot with dual arms uses the invention to rapidly reorganise kitchen items, toss laundry into a basket, or pass tools between hands while performing maintenance tasks.
CLAIMS
1. A system for dynamic object transfer between robotic manipulators, comprising:
- a first robotic manipulator (the throwing manipulator) equipped with at least one force, torque, or tactile sensor;
- a mass and inertial estimation module configured to estimate the mass and inertial properties of a grasped object based on sensor readings from the first robotic manipulator;
- a trajectory computation module configured to compute a release velocity vector and release timing for the object based on the estimated mass and inertial properties, a target intercept region associated with a second robotic manipulator, and gravitational and optionally aerodynamic parameters;
- a second robotic manipulator (the receiving manipulator) equipped with at least one high-speed tracking sensor;
- an in-flight tracking module configured to track the position and velocity of the object after release and to predict an intercept point and arrival time; and
- a catch synchronisation controller configured to command the second robotic manipulator to execute a timed grasp at the predicted intercept point.
2. The system of claim 1, wherein the mass and inertial estimation module performs haptic interrogation comprising at least one of: static weighing, dynamic oscillatory probing, or impulsive motion probing.
3. The system of claim 1, wherein the trajectory computation module propagates uncertainty bounds from the mass estimation through the ballistic model to produce a predicted landing distribution, and wherein the catch synchronisation controller plans a catch envelope covering said distribution.
4. The system of claim 1, further comprising a closed-loop learning module that records transfer outcomes and updates at least one of: mass estimation calibration parameters, ballistic model parameters, or catch timing parameters.
5. The system of claim 1, further comprising a safety module that: (a) verifies the computed trajectory does not intersect a human-occupied zone before authorising the throw; (b) constrains the maximum kinetic energy of any thrown object; and (c) activates a passive catch mechanism upon prediction of a missed catch.
6. The system of claim 1, wherein the first robotic manipulator executes a wrist-rotation or sequential finger-extension release to impart stabilising spin to the object during release.
7. The system of claim 1, wherein the second robotic manipulator velocity-matches its end-effector to the predicted arrival velocity of the object to reduce relative impact velocity.
8. The system of claim 1, wherein the first and second robotic manipulators are arms of a single dual-arm robotic platform.
9. The system of claim 1, wherein the first and second robotic manipulators are separate robotic platforms communicating via a deterministic low-latency network.
10. A method for dynamic object transfer between robotic manipulators, comprising the steps of:
(a) grasping an object with a first robotic manipulator;
(b) estimating the mass and inertial properties of the object using force, torque, or tactile sensor data from the first robotic manipulator;
(c) computing a release velocity vector, release point, and release timing based on the estimated mass, a target intercept region of a second robotic manipulator, and ballistic trajectory parameters;
(d) executing a throwing motion with the first robotic manipulator and releasing the object at the computed release state;
(e) tracking the object in flight using at least one high-speed sensor and predicting an intercept point and arrival time;
(f) commanding the second robotic manipulator to move to the predicted intercept point and execute a timed grasp synchronised to the predicted arrival time.
11. The method of claim 10, further comprising, after step (f), recording the transfer outcome and updating at least one of: mass estimation calibration, ballistic model parameters, or catch timing parameters based on the recorded outcome.
12. The method of claim 10, further comprising, before step (d), verifying that the computed trajectory does not intersect any human-occupied zone or obstacle, and aborting the throw if the verification fails.
13. The method of claim 10, wherein step (b) comprises executing a controlled oscillatory or impulsive motion with the first robotic manipulator and deriving the object’s mass from the ratio of applied force to measured acceleration.
14. The method of claim 10, wherein step (f) further comprises velocity-matching the second robotic manipulator’s end-effector to the object’s predicted arrival velocity.
15. A non-transitory computer-readable medium storing instructions that, when executed by one or more processors, cause the one or more processors to perform the method of claim 10.
ABSTRACT
A system and method for transferring objects between robotic manipulators via adaptive ballistic handover. A throwing manipulator estimates the mass and inertial properties of a grasped object through haptic sensing, computes an optimal ballistic trajectory and release velocity, and executes a throw. A receiving manipulator tracks the object in flight using high-speed sensors, predicts the intercept point and timing, and executes a synchronised catch with velocity matching and adaptive grip force. A closed-loop learning system records transfer outcomes and continuously refines the estimation, trajectory, and catch models. Safety systems prevent throws into human-occupied zones, limit kinetic energy, and provide passive catch mechanisms for missed transfers. The invention enables significantly faster inter-manipulator object transfer compared to static handover methods, with applications in industrial manufacturing, warehouse logistics, and home robotics.
INVENTOR(S)
ASHTAR VENTURA
APPLICANT
ASHTAR VENTURA
DATE
March 7, 2026
