Tangent Robotics Builds Fine Motor Skills for Robots

Tangent Robotics is building the fine motor skill layer for robot hands. The New York City company combines custom hand hardware, light-based touch sensing, teleoperation, and motor learning so robots can perform the contact-rich work that still defeats conventional automation.

Founded in 2022 as a Columbia Engineering spinout, Tangent is led by co-founder and CEO Pedro Piacenza. Co-founder Matei Ciocarlie serves as Chief Science Officer, Ioannis Kymissis is a co-founder and consultant for unconventional electronics, and Tianjian Chen is CTO.

Tangent matters now because industrial robotics has become very good at moving things through space and remains much less convincing at the last few millimeters. Threading, connector mating, gasket seating, gear installation, wiring, and reassembly require touch, adaptation, and fine control. That gap is where Tangent is placing its bet.

The Missing Layer Is Motor Skill

The current physical AI conversation is crowded with foundation models, vision-language-action systems, and general-purpose robot bodies. Tangent is focused on the layer underneath the plan and above the motor controller: the learned skill that turns "insert the connector" into the tiny corrections required to actually insert it.

Tangent's product thesis is that dexterity emerges from the interaction of hand design, sensing, teleoperation, and motor learning. Treating any one of those pieces as a bolt-on creates a familiar robotics problem: an impressive demo held together by calibration rituals and optimism.

The company's target work is industrial and contact-heavy. Tangent points to assembly, threading, part sequencing, connector mating, gasket seating, snap fitting, gear meshing, installation, inspection, maintenance, and wiring. These are not glamorous tasks, but they are exactly where automation economics get real. A robot that cannot feel the moment a part binds is not replacing the technician. It is waiting for one.

Tangent Builds the Hand and the Skill Stack

Tangent's first platform, TR01, is a two-finger hand designed around precise relative fingertip tracking. Instead of copying human anatomy, Tangent designed the kinematics so a human operator's fingertip motion can be mapped into a useful manipulation workspace. The goal is cleaner teleoperation data for training autonomous policies.

The company then added Popcorn, an optical touch sensor mounted at the base of each fingertip. Tangent says the sensor streams 14 raw optical signals at 250 Hz, can detect forces as low as 0.1 newton in realistic conditions, and keeps the replaceable contact surface separate from the sensing electronics. That is a practical choice. Sensors built for the lab tend to become precious. Sensors built for manipulation need to survive being used.

TR02 adds a third finger and a functional palm for more stable grasps, in-hand reorientation, and simultaneous stabilization and manipulation. TR01 Heavy increases fingertip force, improves electronics and cable routing, and adds a lower-latency camera. Together, the systems show a company iterating across mechanics, sensing, perception, and learning rather than betting the business on one heroic model.

The Research Lineage Is Real

Tangent came out of Columbia Engineering's Robotic Manipulation and Mobility Lab. The founding team worked on tactile sensing, dexterous hands, NASA Astrobee mechanisms, and the manipulation stack for ROS.

Columbia documented a tactile finger that used light and machine learning to localize contact across a curved surface. Later work demonstrated a five-finger hand performing in-hand manipulation using touch and proprioception without external cameras. TIME included that research hand in its 2023 Best Inventions list.

That history gives Tangent technical credibility, but academic lineage is not a commercial shortcut. Manufacturing environments care about repeatability, cycle time, changeover, maintenance, safety, and cost. Tangent still has to prove that its integrated approach survives those constraints outside controlled demonstrations.

From Demonstrations to Industrial Applications

Tangent has published autonomous visuotactile demonstrations using imitation learning, vision, Popcorn touch data, and teleoperated examples. The company reports that its policies were trained from dozens of complete demonstrations plus correction demonstrations, with representative tasks including threading, gear installation, and small-object manipulation.

Those are company-reported results, not audited production benchmarks. Tangent also says it is working with pilot commercial partners on skilled assembly and manufacturing applications, but it has not named them. The distinction matters. Robot videos prove motion. Customers prove that the motion is dependable enough to buy.

The $4.5 million pre-seed round, co-led by Fly Ventures and Toyota Ventures, gives Tangent capital to move from research platforms toward demanding manufacturing deployments. Logos Fund and Sparked Ventures also participated. The investor mix fits the work: technically difficult hardware, long validation cycles, and a potential control point in the physical AI stack.

Leadership Spans Robotics Research and Product Building

Pedro Piacenza leads Tangent as CEO. Matei Ciocarlie brings robotics research and prior work on the Willow Garage manipulation team. Ioannis Kymissis adds unconventional-electronics expertise, and Tianjian Chen leads technology development as CTO.

The broader team includes researchers and engineers working across mechanisms, robot learning, software, and manipulation. That mix reflects the product. Dexterity is not only a hardware problem, and it is not only an AI problem. The hand, sensor, control system, data pipeline, and policy all have to agree about what just happened at the fingertip.

Tangent's current New York City openings include a Research Scientist in Robot Learning and a Robotics Software Engineer, plus a general application path. The careers page asks for skills across reinforcement learning, imitation learning, control, kinematics, planning, vision, mechatronics, firmware, simulation, and real-hardware integration.

Hiring across that range is a market signal. Tangent is building the connective tissue required to turn a dexterous hand into an operating system for manipulation. It also reveals the cost of the ambition: full-stack robotics means owning a lot of hard problems at once.

Why Tangent Robotics Matters

The robotics market does not lack intelligence demos. It lacks dependable fine motor skill.

Tangent's strategic edge is the decision to co-design the hardware, touch system, teleoperation interface, and learning stack around contact-rich work. If that loop produces faster skill acquisition and reliable industrial performance, Tangent could become an enabling layer for robot makers and manufacturers trying to automate their most stubborn manual steps.

The company still has major questions to answer. Public evidence does not establish production deployments, customer economics, cycle-time performance, audited reliability, or the first commercial configuration. Those are not footnotes. They are the bridge between an excellent robotics program and a durable company.

But the market timing is sharp. Physical AI is moving from "can the robot understand the task?" to "can it execute the task without dropping the part, jamming the connector, or calling a human every six minutes?" Tangent Robotics is building for that second question, which may be the one that determines whether the first question matters.