Jingshi Technology's VIVA Hand Achieves Real-Time Human-Robot Four-Hand Piano Duet

Jingshi Technology recently released the VIVA dexterous hand, which achieved the world's first real-time human-robot four-hand piano duet. The robot no longer merely plays accurately; it plays expressively.

The company also announced the VIVA × CADA milestone, saying robot music intelligence has entered a stage of creation, performance and continuous learning. Robots are moving from preset motions to artistic expression and creative partnership.

VIVA comes from the Latin root viv, meaning alive, and echoes the musical term vivace. CADA derives from Cadenza, symbolizing a performer's free expression. Together they evoke "Viva cada dia," enjoy every day.

VIVA aims to break the industry's impossible triangle of strength, speed and dexterity from the hardware root.

In dexterous hands, power limits force a tradeoff between strength and speed, while higher degrees of freedom cap each actuator's power. Jingshi Technology used the human hand as its benchmark to solve this through forward design.

The team collected and analyzed data from human pianists, extracted core design metrics, built theoretical models and optimized the mechanical design through topology optimization.

As a result, complex hand shape changes, multi-fingertip coordination and fine trajectory control approach the motion boundaries of a human hand.

The key is a first-of-its-kind tendon-linkage hybrid transmission. Tendon drive places actuators at the rear, reducing distal mass so fingers gain high speed. Mechanical linkages ensure stiffness and precise force transfer for strong fingertip output. The hybrid system reproduces the natural coordination of human joints, making each key press and finger lift follow a human-like trajectory rather than stiff mechanical motion.

A high-performance drive system pushes joint speed beyond 1,000°/s and single-fingertip peak force to 33 N. This covers the full dynamic range, from fast finger changes to forceful accents. Fast scales, trills and repeated notes are stable, providing enough performance for time- and force-sensitive human-robot duets.

Professional piano playing demands extreme strength, speed and dexterity. The difficult Can-Can duet with a human partner was the challenge that drove the effort to solve the dexterous hand's impossible triangle.

VIVA × CADA: VIVA is the dexterous performer; CADA is the music embodied large model.

CADA, Jingshi Technology's self-developed music embodied large model, gives VIVA its "thought." If VIVA is the body that performs music, CADA is the brain that grants musical ability.

CADA closes the loop from music understanding and creation to embodied performance and feedback learning. A general large model interprets natural language instructions and musical intent. A music generation model creates melody and arrangement. A performance strategy model translates abstract expression into fingering, touch, dynamics and timing. Finally, VIVA performs on a real piano.

Every performance is not an endpoint. The system continuously perceives the sound, motion and results, compares goals with actual effects, analyzes errors and adjusts strategy. This forms a creation-performance-perception-evaluation-learning loop, so the same piece improves over time.

The performed "Viva la Vida: Nueva Vida" was generated through interaction between the large model and a human, then autonomously learned and corrected by the robot, expressing its own romance.

Jingshi Technology says VIVA represents more than a piano-playing hand. When it touches keys, controls force, feels melody and finally presents music, the machine begins to express rich emotion.

Jingshi Technology views music as a highly integrated scenario combining cognition, creation, expression and precise operation, making it an ideal carrier for deep integration of large models and robots.

General large models provide understanding, reasoning and interaction. Domain-specific models convert general intelligence into professional skills. The robot body executes in the real world, while continuous environmental feedback lets skills evolve. This is an important direction for future robotics.

Starting from understanding and playing music, the team hopes robots will eventually gain the ability to understand, create, act, interact and continuously learn.

Let machines have a body, let the body emerge thought, and let thought be heard.

The VIVA × CADA milestone moves robots from executing actions toward artistic expression, advancing intelligence from the digital world into the physical world.