Tesla Optimus, Figure 02, and 1X NEO: Same Autonomy Level, Different Trade-offs

Tesla Optimus represents the vertical integration gamble. The company manufactures humanoids while running the FSD fleet and controlling the underlying silicon. The stated logic: if autonomous vehicle perception and planning can be trained across millions of real-world miles, that foundation should transfer to a biped.

Tesla has demonstrated prototype assembly tasks, facility navigation, and object manipulation. Public demonstrations often include teleoperated segments.

That transparency about hand-on-joystick is also, precisely, what defines the autonomy level.

The bet is that iterating FSD’s learned representations and closed-loop planning at scale will eventually enable higher autonomy in humanoid form. The robot is, in effect, a rollout of an existing autonomous platform into a new embodiment.

Level II reflects that reality. Optimus can operate independently in constrained, known environments but requires human monitoring in open-ended scenarios. The investment is long-term. The bet is on momentum and manufacturing scale.

Figure 02 began with a different premise. An initial OpenAI partnership positioned Figure as a consumer of large-scale foundation models: inherit general reasoning and vision-language understanding, then handle embodiment and hardware separately.

That partnership has evolved. Figure now trains Helix, its own vision-language-action model, on on-robot data. The underlying premise holds: use learned representations from large models to reduce the engineering burden for task autonomy.

Figure’s commercial milestone was a pilot on BMW production lines, handling assembly and material-handling tasks. This is not a prototype demo. It is a production trial with real constraints: cycle times, safety regulations, integration with existing systems.

Even so, Figure remains at Level II. The robot executes pre-learned tasks within scripted material flows and relies on environmental structure. Performance on novel or off-script situations has not been demonstrated at production scale.

The bet is that foundation models will generalize. If a robot trained on diverse real-world data can adapt to task variation through language-grounded reasoning, Figure could move up the Autonomy Ladder faster than competitors still engineering task-specific solutions.

1X takes the most unconventional form factor: a soft-bodied humanoid optimized for household tasks. The design philosophy emphasizes safety through compliance rather than force limits, and domestic applicability from the start.

1X has been explicit about its teleop hybrid model. NEO operates via a subscription service where users can teleoperate for certain tasks while the robot handles simpler ones independently. This is not presented as a limitation to overcome.

It is the current product design: a spectrum from autonomous to remote-operated, with human involvement as a feature.

A soft-bodied robot moving through homes, handling fragile objects, and interacting with living spaces faces different constraints than a rigid industrial platform. If the engineering challenge is reduced through compliance, the autonomy development timeline may shorten with it.

Level II reflects capability in structured household scenarios with significant teleop support for variation and exception handling. The company is betting that household robotics does not require the generalization ceiling of industrial humanoids, and that form factor is a faster path to adoption than brute-force autonomy in rigid frames.

All three robots sit at Level II: Assisted Autonomy with monitored execution. Each can operate independently in defined scenarios, but human oversight or intervention is required when the scenario deviates from the known distribution.

The Autonomy Ladder^™ uses a behavioral definition, not a technological one. Level II is not about the presence of a large language model or the elegance of a control architecture. It is about what the robot demonstrates: can it detect when it is outside its operating envelope, and does it gracefully yield to human control?

Level II robots do. They monitor. They pause. They ask for help.

Level III requires something different. A Level III robot operates across diverse, unstructured scenarios without human monitoring, within defined operational boundaries. No teleoperation. No remote joystick. Conditional, but independent.

Tesla Optimus has the infrastructure to approach Level III, but current demonstrations include teleop fallback. Figure 02 has foundation models that generalize, but pilots remain task-constrained and monitored. 1X NEO is explicitly designed with teleop as a product feature, not a gap.

Specific signals would indicate movement up the Autonomy Ladder over the next 12 to 18 months. These are concrete and observable.

For Tesla Optimus: independent operation in open-ended environments without teleop fallback, across tasks not seen in training. A multi-month factory deployment with documented autonomy, not a demo.

For Figure 02: task generalization beyond the BMW pilot. Can Helix-trained Figures operate in different factories, handle different product flows, adapt to new equipment? The foundation model bet only pays off if the robot transfers knowledge across deployments.

For 1X NEO: a measurable reduction in teleop utilization over time within the same household scenarios. If launch sees high teleop rates and six months later the same homes see significantly lower rates at constant task complexity, that is genuine progress. The metric is autonomy within the household domain.

All three companies control the data and the timelines. The question is not theoretical capability ceiling, which may be high for all three. The question is whether the capability exists today in consistent operation.

Comparing these robots at the same autonomy level can appear to miss the story. Three different form factors, three different deployment models, three different technology bets.

They are not competitors on a single metric. They are experiments in different paths to autonomy. Tesla is betting on scale and infrastructure reuse. Figure is betting on foundation models and enterprise deployment. 1X is betting on form factor and the household domain.

Each path has a logic. Each has risks.

The Autonomy Ladder^™ treats them as equivalent because, in early 2026, they are. None has demonstrated independent, adaptable operation across a diversity of scenarios without human oversight. All require either teleoperation, task scripting, or environmental constraint to function.

The sameness is not a judgment. It is a fact. What matters next is execution: which of these bets produces results, and how quickly the robots move beyond monitoring to true independent operation.