Frontier LLM agents demonstrate remarkable capabilities in directly generating actions for robotic manipulation. Through iterative trial and error, they continually improve manipulation ability by updating skills, generating code and using tools. We call this process physical recursive self-improvement, or physical RSI. Rather than further demonstrating the capabilities of frontier LLM agents, this work asks three fundamental questions: What is the origin of physical RSI capabilities? What scaling laws govern this process? Do the resulting improvements reflect overfitting or generalization?
What allows an agent to turn physical feedback into a better policy? Perception helps it interpret outcomes, spatial reasoning relates observations to actions, and coding and harness use allow it to implement and test revisions. These capabilities may play different roles. We ask whether strengths behind a good first attempt also explain improvement after feedback.
We introduce PhysRSI-eval to study these questions under a common protocol. Six systems revise persistent code, tools and notes across 18 manipulation tasks, with frozen model weights and a fixed physical interaction budget. Gemini 4 is reported separately on ten tasks.
An exploratory comparison of four public capability profiles on the 18 tasks associates spatial reasoning most strongly with initial performance, and perception with attained performance and recovery. This pattern motivates the Physical Barrel Hypothesis: the capability that limits an agent may depend on which aspect of improvement we measure. Starting well and learning from failure may require different strengths. These associations suggest candidate bottlenecks, not causal requirements.
PhysRSI-eval: Benchmark Design
PhysRSI-eval makes the three questions measurable by recording both physical attempts and the workspace carried between them. The attempt sequence reveals where an agent starts and how it progresses under a fixed budget. Workspace interventions test what agents retain and whether it transfers.
What Must the Benchmark Distinguish?
A strong first attempt and useful revision are different abilities. We therefore score a first-attempt baseline separately and follow subsequent attempts under a fixed budget. Per-round scores show when gains occur; charged physical steps show what they cost. Reaching a high score, reaching it early and using fewer interactions are separate outcomes.
Success alone does not show whether the agent has accumulated useful experience. Code, tools and notes must remain accessible for intervention. Clearing the workspace tests whether retained experience helps; freezing it on unseen layouts tests whether it remains useful; transferring it to new tasks tests whether it can support a different goal. We report these tests separately from the main profiles in Section 6.
A Controlled Loop of Physical Attempts
Model weights stay fixed, while skills, tools, policy code and memory can change in a persistent workspace. Each revision is tested through a judged physical attempt. This connects workspace changes to an observed outcome and a physical cost.
K0 measures the starting point without a persistent workspace. Each task runs three times through the shared robot tools, and its scores are averaged. At K1, the agent receives a Harness workspace that is empty except for the host’s read-only tools. Each round executes one physical attempt and returns the native score, success flag and failure reason. The agent then revises its workspace with robot actions disabled and commits it for the next round. The main evaluation allows at most a fixed, unspecified number of refinement rounds per task and stops at the first native success.
Both stages share the robot interface. K1 also introduces workspace tools and program execution, so the K0-to-K1 gain mixes interface changes with improvement. We therefore study the Harness rounds as the self-improvement segment. Each round allows 4,200 seconds of reasoning; revision takes up to 1,800 seconds for GPT systems and 600 seconds for Claude systems.
For aggregation, successful tasks retain score 1 after stopping, without further execution. Unsuccessful tasks retain their current scores, including regressions. Extra attempts enter the physical budget but add no scoring rounds. Nine GPT-5.6 Sol Harness entries summarize two observed runs. The three-repeat K0 protocol applies to all six systems.
Tasks Across Difficulty Levels
We evaluate six systems: GPT-6 Astra, GPT-6.1 Sol, GPT-6 Sol, GPT-5.6 Sol, Claude Opus 5.5 and Claude Sonnet 5.5. Each system comprises a model and its runtime configuration, which can also affect performance. In a separate experiment without supplementary hints, Astra solves 13 tasks, compared with 15 in the main evaluation. Results from these conditions are reported separately.
Gemini 4 is reported in a separate partial run covering ten tasks through K20, with six solved. Its reported mean scores are 0.050 at K0 and 0.48 at K10. Its five-axis profile is computed on these ten tasks. Because the task set differs from the main 18-task protocol, it is excluded from the benchmark-based association and prediction analyses.
A Five-Axis Profile of Improvement
To study how physical RSI scales, we need more than an endpoint score. The five-axis profile separates the starting point from gain timing, attained score, physical budget savings and recovery from failure. Each axis measures a different aspect of improvement.
How well does the agent perform on its first attempt? The first-attempt score is the mean K0 score, averaged over three runs per task and then across tasks, including partial progress.
Where Does Physical RSI Capability Come From?
What allows physical feedback to become a useful policy revision? Acting, interpreting an outcome and implementing a change may draw on different capabilities. We use the five-axis profile to ask whether the capabilities associated with a strong first attempt also track effective improvement.
Which Capabilities Track Physical RSI?
Strong coding scores alone do not ensure physical improvement. GPT-5.6 Sol exceeds GPT-6 Astra on LiveBench code generation and completion, yet solves one of the 18 tasks, whereas Astra solves 15.
The exploratory analysis uses six systems on all 18 tasks. We group 17 public benchmark scores into Spatial, Harness, Perception and Coding, then compare their ranks with A1–A5.
The Physical Barrel Hypothesis
The pattern motivates the Physical Barrel Hypothesis: different aspects of improvement may have different capability bottlenecks. Spatial reasoning may matter most for starting well, while perception may matter more for interpreting failures and recovering from them. Under this hypothesis, strong coding cannot always compensate for a weakness elsewhere. Its weak associations do not imply that coding is unnecessary; all systems use code here.
These associations identify candidate bottlenecks that targeted capability interventions can test directly.
These candidate bottlenecks lead to the second question: how can we quantify physical RSI and its scaling with model capabilities and interaction budget?
How Can Physical RSI and Its Scaling Law Be Quantified?
How much can an agent improve, when do its main gains occur, and how much interaction does it need? The five-axis profile turns these questions into measurable outcomes. We first quantify differences among agents, then study how these outcomes scale with model capabilities and physical interaction budget.
Observational Scaling Across Model Capabilities
Can public capability scores anticipate physical improvement before the full evaluation is run? This is the cross-model question of observational scaling. Beyond correlating capabilities with outcomes, a predictor must place an unseen system using only the relationships observed in other systems.
What Can Early Physical Feedback Predict?
Once physical interaction begins, can early progress predict further gains? The six systems reach their high-score windows at different times. The first round reaching 90% of peak gain is K9 for Astra, K7 for GPT-6.1 Sol, K15 for Opus, K13 for Sonnet, and K17 for GPT-6 Sol and GPT-5.6 Sol. Forecasting later rounds tests whether a short trajectory contains useful information about continued improvement.
These analyses address the amount and timing of improvement. We next ask what the gains represent: adaptation to the refinement setting, or experience that remains useful elsewhere?
Does Physical RSI Reflect Overfitting or Generalization?
With fixed model weights, generalization concerns the code, tools and notes the agent retains. Do they encode reusable procedures, or do they become specialized to the refinement setting? We first test whether retained experience helps beyond repeated attempts. We then evaluate frozen workspaces on unseen layouts and transfer them to new tasks.
Does Retained Experience Help Beyond Repeated Attempts?
The remaining conditions withhold individual channels. Withholding grader feedback causes the largest drop, to 5/36 (13.9%), and withholding earlier-round images yields 12/36 (33.3%). Keeping only the workspace and grader feedback yields 17/36 (47.2%), so past images and action records add little once these two channels are kept. The no-history control supports a role for retained experience, but leaves open whether that experience is specific to the refinement setting.
What Survives When Revision Stops?
To test whether the stored artifacts work beyond practice, we freeze Astra workspaces after K6 and test each on five unseen layouts, with no further revision or outcome feedback. Across 12 sources, workspace files alone succeed in 25/60 tests (41.7%). Adding prior raw records gives 23/60 (38.3%). The stored artifacts remain useful under layout changes, while these point estimates show no benefit from exposing the raw history at test time.
Fixed-layout source workspaces reach 27/60 (45.0%) on held-out layouts, compared with the reported 14/16 (88%) on their own layout. The different denominators prevent a matched paired interpretation. Still, the comparison suggests that scene-specific gains and transferable experience can coexist. These tests establish some cross-scene usefulness for Astra on this task subset; they do not establish generalization across all systems.
Can Experience Help on a Different Task?
Traces from five models suggest two forms of reusable experience: executable routines and textual operating knowledge. Sonnet 5.5 and Opus 5.5 increasingly encode control in self-written tools, which generate over 96% of their actions in later rounds. GPT-6 Sol also builds a substantial tool library, but tools generate 67% of actions while the model generates 33% directly. These shares measure tool reliance.
Astra and GPT-6.1 Sol follow a different pattern. Approximately 99% of their actions remain directly generated by the model. They use only a very small number of tools for verification, but their reasoning time decreases across rounds. File-access traces show repeated use of memory, skills and policy notes to recover successful actions and subtask solutions, diagnose failures and guide revisions. These observations suggest that both executable routines and retained operating knowledge can reduce repeated reasoning.
We further find that the workspace and generalization analyses align with the Physical RSI Barrel Hypothesis across its four capability factors: spatial reasoning, perception, coding and harness operation. Spatial tools account for 14.5% of Sonnet’s tool calls and 13.7% of Opus’s, with 408 of 2,819 calls and 183 of 1,332 calls, respectively. Astra makes only seven tool calls, including one to such a tool, or 14%. Astra’s minimal reliance on these modules suggests stronger intrinsic perceptual and spatial capabilities. Sonnet and Opus repeatedly reuse these modules across tasks, suggesting that they rely more on tools to compensate for limitations in these capabilities.
During cross-task transfer, agents also consult memory and skills from semantically similar tasks. In Organize Table, Opus combines a parameterized move tool from Pack Objects Into Box with a memory from Store Tools In Toolbox: “the PITCHED 30 deg grasp… reaches the +y half…” The tool supplies an operation, while the memory identifies a grasp orientation that extends its reach. This shows how a reusable procedure and knowledge of when it works can jointly support transfer.
The transfer traces suggest that performance gains depend on experience relevant to the target task’s bottleneck and the agent’s ability to assess its applicability from failure evidence. Semantic similarity helps identify candidate experience; failure evidence guides whether to reuse, adapt or discard it. The teaser illustrates this mechanism through the reuse of a closed-loop grasp-and-transfer routine from stacking blocks to covering blocks with a cup, and an insertion routine from key insertion to charger plugging.
Limitations
The main comparison covers six model-plus-configuration systems, with one improvement chain per task. K0 repeats do not estimate variability across complete chains. First-success stopping records one success; reliable repeated execution needs a separate test. Budget and Improvement use conditional task sets that differ across systems. Astra’s outcomes differ between the main evaluation and the experiment without supplementary hints.
The capability analysis relies on public benchmark scores whose evaluation settings differ from our runs, and it covers six systems without independent validation. Capability bottlenecks and gain-timing predictions therefore remain hypotheses to test. The ablation and frozen tests cover Astra and six tasks, while cross-task results are three first-run cases. These limits constrain the scope of the observed transfer and the mechanisms that can be inferred from it.
Conclusion
PhysRSI-eval studies the origins, measurement, scaling and generalization of physical RSI under a fixed interaction budget. Capability associations motivate the Physical Barrel Hypothesis: the strengths needed to start well may differ from those needed to improve. The five-axis metric makes these outcomes explicit and shows that high scores, early gains and efficient interaction need not coincide. Cross-model prediction and early-round forecasting probe how these outcomes vary with capabilities and interaction budget. Workspace interventions support a role for retained experience and show that it can remain useful on new layouts and tasks. Physical RSI can thus involve both scene-specific adaptation and reu…
Frequently Asked Questions
What is physical recursive self-improvement? It is the process by which an agent improves its manipulation ability through iterative trial and error, updating skills, code and tools.
What is PhysRSI-eval designed to measure? It measures initial performance and improvement across physical attempts, while also testing what agents retain and whether it transfers.
What does the Physical Barrel Hypothesis propose? Different aspects of physical improvement may have different capability bottlenecks; spatial reasoning may matter more for starting well, while perception may matter more for interpreting failures and recovery.
Does retained experience generalize beyond repeated attempts? In tests reported here, frozen Astra workspaces remained useful on unseen layouts, and traces from cross-task transfer showed reuse of tools and operating knowledge.