What we’re watching next in humanoids

Humanoid demos in Tokyo actually move—no vaporware, just motion, today’s headline on the TechCrunch-led tour.

TechCrunch is shipping its Startup Battlefield presence to Tokyo, and SusHi Tech 2026 is sharpening its lens on AI, robotics, resilience, and entertainment. The stated plan is to stage live humanoid demonstrations alongside panels about autonomous software, cyber resilience, and the broader speculative future of AI-inflected media. In other words: a showcase that blends investor-level fascination with hands-on hardware, where a dozen cameras tend to reveal more about a robot’s joints than a PR keynote ever could.

From the material published, no specific robot models or DOF (degrees of freedom) tallies were disclosed. The organizers’ emphasis is on live demonstrations of humanoids, indicating a lean toward what you can observe in real time rather than what marketers tell you in slides. That matters. In the humanoid space, “live demo” is a strong signal about engineering discipline: can the robot stand upright, balance when nudged, grab, and respond to simple tasks without glitching into a choreography of jitter? The absence of granular metrics in the initial coverage also highlights a perennial risk in the field: demonstrations can be staged to look convincing, but the underlying control software, perception stack, and actuation reliability are where the real readiness lives.

Engineering realities underlie the excitement. A humanoid’s ability to walk, manipulate objects, and respond to dynamic environments hinges on a cluster of tough, interdependent constraints: joint torque, control bandwidth, energy management, perception fidelity, and safety frameworks. The fact that SusHi Tech 2026 explicitly includes robotics in a broader media-tech context signals that the industry is moving beyond “one trick” demos toward integrated systems—where sensing, control, and manipulation must operate in concert. The absence of disclosed payload or joint-count numbers means engineers, investors, and operators must read the demos with caution: a convincing walk may still rely on tethered power, or a tightly choreographed path that glosses over real-world variances such as uneven terrain, lighting, or unexpected tactile feedback.

From a practitioner’s perspective, the most meaningful takeaway will be the cadence between demonstration and deployable capability. The field continues to wrestle with two critical realities: energy density and autonomous decision-making in unstructured settings. Most humanoids in late-stage prototyping feature tens of joints (often 20–40 DOF overall) and a range of hands/grippers designed for everyday objects rather than industrial payloads. Arm and hand configurations commonly prioritize dexterity for grasping a variety of items, while leg mechanisms balance stability and energy efficiency. That translates to a practical limitation: even when a robot can stand, walk, or lift a light object, it may not yet sustain longer runtimes or operate safely without depth-perception and tactile feedback loops that are still maturing.

At the same time, the event’s framing around AI-enabled software and autonomous driving innovations creates an environment where the hardware is judged alongside software reliability. Demonstration footage shows the hardware moving, but the real question is whether the software stack can generalize across tasks, adapt to new objects, and maintain safety margins in real time. In other words, does the robot’s autonomy rely on brittle, choreographed routines, or on robust perception-control loops that survive real-world noise?

What’s clear: the tech is not yet “field-ready” for broad deployment, but the demonstrations will provide sharper signals about what’s next. Expect a wave of follow-up disclosures from participating teams—DOF tallies, exact payloads, battery runtimes, and safety certifications—before the next wave of investment decisions.

What we’re watching next in humanoids

DOF and payload transparency: will organizers publish joint counts, hand dexterity specs, and object payload limits for the showcased models?

Runtime and power: how long can the demos run on battery, and what charging regimes are implied for day-long clinics or facilities?

Autonomy in the wild: will demos show independent task execution beyond scripted sequences, and how do perception and planning hold up under real-world variability?

Safety and compliance: what standards and test protocols accompany demonstrations, and how quickly will that translate into field deployments?

Software stack disclosures: which control frameworks, perception engines, and AI back-ends power the movements, and how portable are they across platforms?

Sources

TechCrunch is heading to Tokyo — and bringing the Startup Battlefield with it