Embodied First Exam
Embodied First Exam
The authors introduce the Embodied First Exam, a comprehensive evaluation framework designed to assess coding agents' ability to control diverse robots under physical-world constraints. The framework supports three control approaches: direct control, code-to-control, and policy training. It includes 1,156 tasks across 35 suites on 47 robots, covering spatial puzzles, dexterous manipulation, multi-robot interaction, dynamic whole-body control, and sim-to-sim engineering. Tasks can be evaluated in Standard Setting (sensor-only, no environment reset) or Privileged Setting (full simulator state access). On 456 core tasks, GPT-6 Astra in Codex achieves a 34% success rate in the Standard Setting.
The paper introduces the Embodied First Exam, a unified evaluation framework for coding agents that supports three control modes (direct, code-based, and policy training) across 47 robots and 1,156 tasks in 35 suites, with results reported for Standard and Privileged settings to distinguish real-world constraints from oracle upper bounds.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
tasks
1,156
View original evidence
, and agent-native evaluation framework where coding agents can choose their own control approach: direct control, code to control, or train a policy to control. We hope it serves as a first exam for coding agents before they enter the physical world, with 1,156 tasks in 35 suites on 47 robots, including 9 new suites that go beyond pick-and-place to spatial puzzles, dexterous manipulation, multi-robot interaction, dy
Open source S3suites
35
View original evidence
, and agent-native evaluation framework where coding agents can choose their own control approach: direct control, code to control, or train a policy to control. We hope it serves as a first exam for coding agents before they enter the physical world, with 1,156 tasks in 35 suites on 47 robots, including 9 new suites that go beyond pick-and-place to spatial puzzles, dexterous manipulation, multi-robot interaction, dy
Open source S3Robot count
47 robots
View original evidence
, and agent-native evaluation framework where coding agents can choose their own control approach: direct control, code to control, or train a policy to control. We hope it serves as a first exam for coding agents before they enter the physical world, with 1,156 tasks in 35 suites on 47 robots, including 9 new suites that go beyond pick-and-place to spatial puzzles, dexterous manipulation, multi-robot interaction, dy
Open source S3core tasks
456
View original evidence
lowing it to search for an oracle solution for training purposes. Success rates in this setting serve as an approximate upper bound. On the 456 core tasks, the strongest of eight agents, GPT-6 Astra in Codex, achieves a success rate of 34% in the Standard Setting, leaving much room for improvement. To see where agents fail, we label the stage at which failed attempts go wrong and compare failures across agents, setti
Open source S5Reported success rate
34%
View original evidence
lowing it to search for an oracle solution for training purposes. Success rates in this setting serve as an approximate upper bound. On the 456 core tasks, the strongest of eight agents, GPT-6 Astra in Codex, achieves a success rate of 34% in the Standard Setting, leaving much room for improvement. To see where agents fail, we label the stage at which failed attempts go wrong and compare failures across agents, setti
Open source S5
- paper: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Dataset licensing and commercial-use rights are not established here.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
, and agent-native evaluation framework where coding agents can choose their own control approach: direct control, code to control, or train a policy to control. We hope it serves as a first exam for coding agents before they enter the physical world, with 1,156 tasks in 35 suites on 47 robots, including 9 new suites that go beyond pick-and-place to spatial puzzles, dexterous manipulation, multi-robot interaction, dy
Open source S3
lowing it to search for an oracle solution for training purposes. Success rates in this setting serve as an approximate upper bound. On the 456 core tasks, the strongest of eight agents, GPT-6 Astra in Codex, achieves a success rate of 34% in the Standard Setting, leaving much room for improvement. To see where agents fail, we label the stage at which failed attempts go wrong and compare failures across agents, setti
Open source S5
Implications for data suppliers
RoboSignal interpretation and collection questions, not statements of buyer demand.
- Confirm the required data type and collection setting with the buyer; this source does not establish a complete collection specification.
- Compare the reported units and scope before using these quantities in a budget. Recording hours, sensor-hours and trajectories are different measures.
- Validate demand and acceptance criteria with a buyer before scaling. Publication, popularity and a research result do not establish a purchase commitment.
Source：arXiv Robotics — research abstracts · arxiv.org