2026 Physical AI Trends: Tactile Sensing and Deformable Material Control
Physical AI is converging on two hard problems in 2026: giving robots a sense of touch and teaching them to handle soft, deformable materials reliably.
5 min read
0:00
0:00
Physical AI is converging on two hard problems in 2026: giving robots a sense of touch and teaching them to handle soft, deformable materials reliably.
Three independent developments in April and May 2026 all point to the same gap: robots still lack the sensory and mechanical resolution to handle the physical world reliably.
From a builder's perspective, the most useful signal in a trend is when unrelated teams converge on the same problem. In the past two weeks, a Hong Kong-based sensor company released a massive tactile dataset, German university researchers published work on light-controlled artificial muscles, and a manufacturing software CEO argued that deformable materials represent physical AI's real manufacturing test. These are not coordinated announcements. They are independent signals pointing at the same bottleneck: physical dexterity. The specs and the use cases are finally starting to align, but the gap between lab capability and factory deployment is still wide.
DAIMON's fingertip sensor packs over 110,000 effective sensing units into a module the size of a fingertip, and their dataset is built to train models on that resolution at scale.
According to IEEE Spectrum, DAIMON Robotics is a two-and-a-half-year-old company that has built its identity around a monochromatic, vision-based tactile sensor. The headline number is 110,000 effective sensing units in a fingertip-sized module. That density matters because most robot hands today operate with much coarser tactile feedback, if they have any at all. The Daimon-Infinity dataset was released in April 2026 and is described as omni-modal, meaning it combines tactile, visual, and other sensor streams. The collection network is designed to generate millions of hours of data annually, which is the kind of scale needed to train physical AI models that generalize across tasks. Partners include Google DeepMind, Northwestern University, and the National University of Singapore.
110,000 sensing units per fingertip is not an incremental improvement over existing tactile sensors. It is closer to human fingertip sensitivity, which has roughly 2,500 mechanoreceptors per square centimeter in the most sensitive areas. Closing that gap changes what force control algorithms can actually do. Whether DAIMON's sensor holds up under repeated mechanical stress in real deployments is still an open question, but the resolution number alone represents a meaningful shift in what is technically available.
Researchers at Friedrich-Alexander-Universitat Erlangen-Nurnberg are scaling individual molecular machines into 3D structures that contract and expand under light, creating a new class of soft actuator.
As reported by Interesting Engineering, researchers at FAU in Germany are working on shape-shifting materials that function as artificial muscles, controlled by light rather than electrical current or hydraulic pressure. The approach scales molecular machines into three-dimensional structures. From an actuator design perspective, this is interesting because it sidesteps some of the fundamental trade-offs in conventional actuators: no gearbox friction, no backdrivability issues from gear reduction, and potentially high force-to-weight ratios depending on material density. These are early-stage research results, not production-ready components. But the direction is consistent with a broader push in physical AI toward softer, more compliant actuation that can interact safely with humans and deformable objects.
Backdrivability is one of the properties most discussed in humanoid robot actuator design. A backdrivable actuator allows external forces to move the joint, which is critical for safe human interaction and for handling objects that push back unpredictably. Conventional harmonic drives are not backdrivable. Series elastic actuators add compliance mechanically. Light-controlled soft materials could offer intrinsic compliance without added mechanical complexity, though the force output and cycle life at industrial duty cycles remain unproven at this stage.
Unlike rigid parts, deformable materials like fabric and flexible components change shape unpredictably during handling, requiring force control and real-time sensory feedback that current systems struggle to provide consistently.
According to The Robot Report, Createme's CEO has argued that physical AI can tackle the assembly of deformable materials in apparel and beyond, with the right software approach, and frames this as physical AI's real manufacturing test. The argument is that software and system design matter as much as hardware capability when handling materials that do not hold a fixed shape. Apparel manufacturing is the canonical example: fabric folds, stretches, and shifts during every step of the handling process. A robot that can pick and place rigid parts with high repeatability can fail completely on the same task when the part is soft. This is not a new problem in robotics, but the framing around physical AI raises the stakes because the entire value proposition of deploying general-purpose robots in manufacturing depends on solving it.
Tactile sensing at scale, soft actuation research, and deformable material handling challenges all converge on the same root problem: physical AI systems lack the sensory and mechanical resolution to match human dexterity.
Mapping these three signals together produces a clearer picture than any single source offers on its own. DAIMON's 110,000-unit tactile sensor and dataset addresses the sensing side of the dexterity problem. FAU's light-controlled artificial muscles address the actuation side, specifically the compliance and force resolution gap. Createme's framing of deformable materials as the real manufacturing test identifies where these gaps cost real money in real deployments today. The pattern is a field converging on a hard problem from multiple directions simultaneously. Institutional partners like Google DeepMind working alongside hardware startups like DAIMON, while university labs explore fundamentally different actuation physics, suggests the investment and research attention is concentrated in the right place.
Watch tactile sensor deployment in real manufacturing environments, soft actuator research moving from lab to prototype, and which companies build the data infrastructure to train on high-resolution physical feedback.
The three trends covered here are at very different stages of maturity. DAIMON's tactile sensor and dataset are shipping and operational in 2026, with a named partner network. FAU's light-controlled artificial muscles are at the research publication stage, with no commercial timeline confirmed. Createme's framing of the deformable material problem reflects active commercial deployment experience. For engineers evaluating actuator and sensor choices, the tactile resolution number from DAIMON is the most immediately actionable data point. For investors and founders, the more strategic question is which companies are building training data infrastructure alongside hardware, because that combination is what enables generalizable physical AI rather than narrow task-specific automation.
According to IEEE Spectrum, Daimon-Infinity is described as the largest omni-modal robotic dataset for physical AI, released in April 2026. It combines high-resolution tactile sensing data with other modalities and covers tasks from home laundry folding to factory assembly line work.
As reported by Interesting Engineering, FAU researchers are scaling molecular machines into 3D structures that respond to light rather than electrical current. This approach could offer intrinsic mechanical compliance without gearboxes, potentially addressing backdrivability and force resolution trade-offs in conventional actuator designs.
According to The Robot Report, deformable materials like fabric change shape unpredictably during handling. Unlike rigid parts with fixed geometry, soft materials require continuous real-time force control and tactile feedback. Current physical AI systems struggle to maintain reliable grip and manipulation across these variable conditions.
That density approaches human-level tactile resolution. Most robot hands today operate with coarser feedback or none at all. Higher sensing resolution enables more precise force control algorithms, which directly affects a robot's ability to handle delicate or deformable objects without damaging them or dropping them.
According to IEEE Spectrum, DAIMON Robotics' Daimon-Infinity project includes collaborative support from Google DeepMind, Northwestern University, and the National University of Singapore, representing a combination of industry AI research and academic robotics expertise.