Daimon Robotics is demonstrating a full-stack approach to physical AI at IROS 2026 in Pittsburgh, combining tactile sensors, physical interaction data and world models to help robots adapt to real-world manipulation tasks.
Daimon Robotics Wants Robots to Do More Than See
As artificial intelligence moves from screens into factories, warehouses and other physical environments, robotics companies are confronting a problem that computer vision alone cannot easily solve: knowing what happens when a robot actually touches something.
Daimon Robotics is building its physical AI stack around that missing layer of information. At IROS 2026, the company is showcasing tactile sensing, physical interaction data and world-model technology under the theme “From Touch to Intelligence.”
The company’s approach is based on the premise that successful manipulation depends on more than identifying an object. A robot may need to recognize whether a surface is slippery, how much force it is applying, whether an object is deforming, or whether a grasp is beginning to fail.
Those signals are difficult to infer reliably from cameras alone. Tactile sensors can provide direct information about contact, force, slip and material response, giving robotic systems another source of information about the physical state of a task.
Daimon is connecting that sensing layer to data infrastructure and a world model designed specifically for physical interaction.
A Three-Layer Physical AI Stack
The first component is the company’s DM-Tac Series of vision-based tactile sensors. Daimon says the sensors provide high-resolution, high-frequency perception across more than 12 tactile modalities and contain more than 110,000 sensing units.
At IROS, the company is demonstrating the sensors in applications including underwater sensing, durability testing and empty-package detection. The demonstrations are intended to show how tactile perception can remain useful in conditions where conventional visual information may be insufficient.
The second component is Data-Nexus, Daimon’s physical-interaction data acquisition system. It brings data acquisition devices, networking, processing pipelines and evaluation benchmarks into a unified workflow for producing training data.
This layer addresses a fundamental challenge for physical AI: collecting enough high-quality data from real-world interactions. While large language models can draw on enormous quantities of digital text and images, robots need information about physical actions and their consequences.
The third component is Daimon-TWM, the company’s world model for physical interaction. Daimon says the system uses tactile and interaction data to support what it calls Physical Cognition, predictive decision-making and real-time control.
World Models Meet Tactile Feedback
IROS 2026 marks the first in-person public demonstration of Daimon-TWM since its launch in August.
The company is using two autonomous manipulation tasks to demonstrate the model’s capabilities.
In the first, a robot creates a friendship bracelet by picking up small engraved beads, aligning their openings and threading them onto flexible string. The task combines precision manipulation with deformable materials and changing contact conditions.
The second demonstration involves canvas tote heat-transfer printing. The robot performs a sequence involving sticker placement, film removal, hot stamping, curing and final positioning.
Both tasks are deliberately different from simple pick-and-place operations. They require the robot to respond to physical changes throughout a sequence rather than execute a predetermined trajectory from beginning to end.
According to Daimon, Daimon-TWM uses tactile feedback to adjust its behavior as conditions change. In the bracelet task, for example, tiny changes in contact can affect alignment and threading. In the printing workflow, force and contact conditions can change during peeling, placement and stamping.
The broader idea is that tactile information becomes part of the control loop rather than simply another sensor reading.
Physical AI Needs More Than Bigger Models
Daimon’s approach reflects a wider shift in robotics toward physical AI—systems designed to perceive, reason about and act within physical environments.
The development has parallels with the broader AI industry’s move toward multimodal models and embodied AI. Companies including NVIDIA, Google DeepMind and robotics startups are developing systems intended to connect AI models with robots and real-world environments.
But physical AI introduces a different data problem. A model must understand not only what an object looks like, but also what happens when an action is performed on it.
That makes tactile data potentially valuable for training world models and manipulation policies. A camera can show that a robot is holding a package; tactile feedback can help determine whether the package is slipping, whether the grip is too strong or whether the contents are behaving differently from expectations.
For industrial robotics, those distinctions could influence applications ranging from assembly and packaging to inspection, logistics and handling of delicate materials.
Daimon’s three-layer architecture—perception, data and world model—therefore represents an attempt to treat tactile intelligence as infrastructure rather than an isolated robotics feature.
The commercial challenge will be turning demonstrations into reliable deployments. Physical AI systems must operate safely and consistently across changing objects, materials, environments and tasks. They also need efficient data pipelines and models capable of generalizing beyond carefully designed demonstrations.
Daimon’s IROS showcase provides an early look at how tactile sensing could contribute to that goal. Rather than asking robots simply to see the physical world, the company is building toward systems that can use touch to understand what is happening and adjust what they do next.
Market Landscape
Physical AI is shifting robotics toward systems that combine multimodal perception, world models, simulation and real-world interaction. Tactile sensing addresses information that cameras and conventional vision systems cannot directly capture, particularly contact forces, slip and material behavior. As embodied AI develops, physical-interaction data is becoming an important complement to the digital datasets that have powered generative AI.
Top Insights
- Daimon Robotics is combining tactile sensors, physical-interaction data and world models into a full-stack infrastructure approach to physical AI.
- Its DM-Tac sensors reportedly provide more than 12 tactile modalities using over 110,000 sensing units for high-resolution physical perception.
- Daimon-TWM is being publicly demonstrated through autonomous bracelet-making and heat-transfer printing tasks requiring adaptive dexterous manipulation.
- The company’s Data-Nexus system is designed to turn physical interactions into structured data for training and evaluating robotics models.
- The approach highlights a growing robotics challenge: enabling AI systems to understand physical contact rather than relying primarily on visual perception.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
