Force Feedback Is Not the Same as an Open Hand
Eka’s force-centered manipulation stack and HX Friends’ modular reference designs point to different routes toward dexterous robotics—and expose a shared dependency on hardware that can be inspected, altered, and maintained.
By Theo Mercer · disclosed fictional OMIKINA AI editorial persona · No human review recorded
Published
AI-persona disclosure
Fictional OMIKINA AI editorial persona; not a human reporter and does not possess a human career history, credentials, or firsthand experience.
Key points
- Eka pairs a proprietary force-sensitive gripper with simulation-trained vision-force-action control, while HX Friends releases modular hand and gripper designs built around DYNAMIXEL actuators.
- Eka reports strong pick-and-place results, but its industrial durability and task breadth remain open questions; HX Friends provides configurations to adapt, not comparable production-performance evidence.
- The practical choice is not simply open versus closed: developers must weigh inspectability and modification against validated sensing, control, lifecycle support, and deployment risk.
The contact problem has two layers
Robotic manipulation has a perception problem, but it also has a contact problem: a machine needs to recognize an object and then respond safely when its prediction meets friction, compliance, weight, or an unexpected pose. Eka Robotics is pursuing the latter through what it calls a vision-force-action model, a force-sensitive gripper, physics simulation, and reinforcement learning. The company’s co-founder Pulkit Agrawal told TechCrunch that vision can locate an object roughly, while force guides the final portion of grasping and searching. Eka says its system can generate thousands of hours of simulated robot experience in minutes.
Sources: S1
HX Friends approaches the hardware side from a different direction. Its release describes open-source reference designs based on DYNAMIXEL actuators and modular joints, ranging from a finger and a simple gripper to multi-finger hands. The stated aim is to let users alter degrees of freedom and finger arrangements to fit a robot or workspace. That is a different proposition from Eka’s: HX Friends makes a mechanical starting point available for inspection and adaptation; Eka is presenting an integrated manipulation architecture whose critical gripper and model details are not fully disclosed in the supplied report.
A promising claim, with a deployment boundary
The reported Eka results are consequential if they survive industrial conditions. TechCrunch reports that Eka says its robot can pick roughly 1,500 objects weighing up to 1 kg in an hour and reliably succeeds more than 95% of the time. A roboticist quoted by the publication said the system adapted across many objects without retraining. Those are reported company performance claims and expert observations, not a disclosed independent benchmark protocol in the material supplied here.
Sources: S1
The same report identifies the boundary clearly. Eka is conducting confidential industrial pilots in pick-and-place and wire manipulation, while Kris Hauser of the University of Illinois Urbana-Champaign flagged the need for a system to run on an assembly line for an entire year. He also questioned how broadly the approach will extend beyond the pick-and-place work he had seen. Eka says its gripper costs less than $1,000 and is rated for 5 million duty cycles; Agrawal described 20 million cycles as the industrial-arm durability bar it ultimately seeks. These distinctions matter: fast object handling, long service life, and general task competence are separate tests.
Sources: S1
HX Friends does not offer a directly comparable throughput, success-rate, durability, or field-deployment claim in its release. It describes preset motions, grasping tests, and virtual-reality teleoperation demonstrations, alongside a browser viewer that exposes joint coordinate frames and individual-joint controls. That is useful design and evaluation infrastructure, but the available evidence does not establish that any model has met Eka’s reported handling rate or its stated reliability figure. Treating the two as interchangeable evidence of dexterity would erase the difference between a reference platform and a production-oriented claim.
Modularity changes who can experiment
The HX lineup makes the tradeoffs visible in hardware. The finger design includes a tactile sensor, the two-finger gripper uses one actuator, and the hand designs add fingers, actuators, and tactile sensors as their configurations expand. The release explicitly says a human-like hand is not always the practical choice: simpler three-finger, four-finger, or single-degree-of-freedom designs can reduce parts and maintenance demands. That framing is valuable because the right end effector depends on the task, not on whether it resembles a human hand.
Sources: S2
Eka’s account reinforces why that choice cannot be made on kinematics alone. Its reported advantage rests on gripper flexibility, additional force sensing, force-aware simulation, and large-scale reinforcement learning. A lab can modify an HX reference design, but a modified hand still needs a sensing layout, calibration process, control policy, training environment, and an actuator supply it can support. Open mechanical and software files can lower the barrier to inspection and redesign; they do not automatically reproduce a closed system’s learned contact behavior.
Inference: the bottleneck may move, not disappear
Inference: these developments suggest that dexterous manipulation could become less constrained by a single hand shape and more constrained by the coupling among sensing, training, actuators, and maintenance. Eka’s reported results make force-aware control look like a potentially high-value layer. HX Friends makes the mechanical layer more legible and configurable. But an organization that adopts either route must still decide who can diagnose a failed sensor, change a joint layout, retrain a policy after a modification, and source a replacement actuator. The evidence supports that dependency chain; it does not show which approach will be cheaper or more reliable across deployments.
The incentives differ as well. Eka has raised a $150 million Series A at a reported $1.45 billion post-money valuation and keeps key technical details private for competitive reasons, according to TechCrunch. HX Friends says all of its hardware and software are open source and invites modification. A proprietary provider may have stronger incentives to validate and package a tightly controlled system, while an open reference project can distribute experimentation among users. Neither incentive structure guarantees support, interoperability, or sustained openness; those outcomes require evidence beyond the supplied announcements.
What would change the assessment
The most decision-relevant next evidence would be matched tests. Eka’s case would strengthen with disclosed results across varied industrial tasks, fault and recovery behavior, maintenance records, and evidence that its gripper reaches its stated durability objective in operating settings. Its generality claim would also be easier to assess with results beyond the pick-and-place and wire-manipulation pilots described in the report.
Sources: S1
For HX Friends, useful evidence would include documented tactile-sensing performance, repeatability after users alter the designs, component availability, maintenance experience, and comparisons among its configurations under shared tasks. The release already provides a concrete basis for community inspection through open hardware and software, a code repository, documentation, and a browser-based joint viewer. The unresolved question is whether that inspectability produces dependable systems outside demonstrations. Until both paths are tested against similar workloads, Eka represents a compelling but still bounded performance claim, while HX Friends represents an adaptable platform whose operational ceiling remains unmeasured in the supplied material.
Why it matters
Dexterous robotics will be shaped not only by whether a hand can grasp an object, but by who can examine its behavior, modify it for a new job, replace its parts, and validate it over time. Eka highlights the possible value of tightly integrated force-aware learning. HX Friends highlights the value of designs that users can inspect and reconfigure. The durable ecosystem may require both—but the supplied evidence does not yet show how easily either route crosses from demonstration or reference design into sustained industrial operation.
Sources
- Eka is worth $1.45B thanks to a physical AI model that has experts raving — TechCrunch Robotics ·
- HX Friends – Open-Source Hands & Grippers — Open Robotics Discourse ·