WHAT ACTUALLY RAN ON A REAL ROBOT

Real-Robot Evidence Database

See what Physical AI models actually demonstrated on real robots, including task, control mode, trials, failures, interventions and evidence limits.

5 retained recordsLast verified: August 13, 2026Primary-source fields preserved per record

Autonomy

5 of 5 records shownFilters are local and do not create crawlable URLs

TurboVLA

Research result

Model: TurboVLA
Robot: AgileX Piper
Task: Four tabletop tasks: roller, playing card, stapler, bowl stacking
Environment: Lab tabletop
Control mode: Autonomous policy execution after teleoperated training data collection
Autonomy level: Task policy
Trials: 160 total
Success: 87.5% aggregate across the four 40-trial task sets
Failures: 20 total inferred exactly from published trial counts and success rates
Human interventions: Runtime intervention count not reported
Limitations: Controlled tabletop tasks on one arm; no long-duration deployment evidence
Source: TurboVLA project

XR-1

Research result

Model: XR-1
Robot: Six robot embodiments
Task: >120 manipulation tasks
Environment: Real-world research evaluations
Control mode: Learned VLA policy
Autonomy level: Autonomous evaluation under paper protocol
Trials: >14,000 rollouts reported
Success: Not reduced to one cross-task figure here
Failures: Model/task-specific; no single aggregate count represented here
Human interventions: See paper protocol
Limitations: Published research evaluation, not proof of unsupervised production deployment
Source: XR-1 project

RDT-1B

Research result

Model: RDT-1B
Robot: ALOHA dual-arm and other manipulation platforms
Task: Bimanual manipulation, language following, unseen objects/scenes and few-shot skills
Environment: Real robot research evaluation
Control mode: Diffusion action policy
Autonomy level: Autonomous policy evaluation under paper protocol
Trials: See paper
Success: Not collapsed into one universal number
Failures: See task-specific tables
Human interventions: Not normalized here
Limitations: Research task protocols differ from deployment workloads
Source: RDT-1B project

Gemini Robotics 2

Official demo

Model: Gemini Robotics 2
Robot: Apptronik Apollo
Task: Whole-body locomotion plus manipulation demonstrations
Environment: Real robot demonstrations
Control mode: VLA whole-body control
Autonomy level: Official autonomous task demonstrations; exact supervision varies by experiment
Trials: Not disclosed as one common total
Success: Task-specific official results
Failures: Not disclosed as one common total
Human interventions: Not disclosed as one common rate
Limitations: Closed model and controlled evaluation; public data does not establish fleet reliability
Source: Gemini Robotics 2

OpenVLA 7B

Research result

Model: OpenVLA 7B
Robot: Multiple manipulation robots
Task: Language-conditioned manipulation
Environment: Real robot research evaluations
Control mode: VLA policy
Autonomy level: Autonomous policy evaluation under project protocol
Trials: Protocol-dependent
Success: See individual project evaluations
Failures: Protocol-dependent
Human interventions: Not normalized across platforms
Limitations: Different robots require adaptation; benchmark and real-world protocols are not interchangeable
Source: OpenVLA project

Model	Robot	Task	Environment	Control mode	Autonomy level	Trials	Success	Failures	Human interventions	Limitations	Source
TurboVLA	AgileX Piper	Four tabletop tasks: roller, playing card, stapler, bowl stacking	Lab tabletop	Autonomous policy execution after teleoperated training data collection	Task policy	160 total	87.5% aggregate across the four 40-trial task sets	20 total inferred exactly from published trial counts and success rates	Runtime intervention count not reported	Controlled tabletop tasks on one arm; no long-duration deployment evidence	TurboVLA project
XR-1	Six robot embodiments	>120 manipulation tasks	Real-world research evaluations	Learned VLA policy	Autonomous evaluation under paper protocol	>14,000 rollouts reported	Not reduced to one cross-task figure here	Model/task-specific; no single aggregate count represented here	See paper protocol	Published research evaluation, not proof of unsupervised production deployment	XR-1 project
RDT-1B	ALOHA dual-arm and other manipulation platforms	Bimanual manipulation, language following, unseen objects/scenes and few-shot skills	Real robot research evaluation	Diffusion action policy	Autonomous policy evaluation under paper protocol	See paper	Not collapsed into one universal number	See task-specific tables	Not normalized here	Research task protocols differ from deployment workloads	RDT-1B project
Gemini Robotics 2	Apptronik Apollo	Whole-body locomotion plus manipulation demonstrations	Real robot demonstrations	VLA whole-body control	Official autonomous task demonstrations; exact supervision varies by experiment	Not disclosed as one common total	Task-specific official results	Not disclosed as one common total	Not disclosed as one common rate	Closed model and controlled evaluation; public data does not establish fleet reliability	Gemini Robotics 2
OpenVLA 7B	Multiple manipulation robots	Language-conditioned manipulation	Real robot research evaluations	VLA policy	Autonomous policy evaluation under project protocol	Protocol-dependent	See individual project evaluations	Protocol-dependent	Not normalized across platforms	Different robots require adaptation; benchmark and real-world protocols are not interchangeable	OpenVLA project

Methodology

Scripted demos, teleoperation, shared autonomy, autonomous policy tests and deployments are kept separate. Training teleoperation is not mislabeled as runtime teleoperation, and no system is called fully autonomous without explicit evidence.

Included

Published model, robot, benchmark, deployment or failure evidence that can be tied to a retained source.

Not inferred

Unknown autonomy, latency, trial counts, deployment scale and failure causes remain unknown when the source does not disclose them.

Update policy

Versions and historical records remain distinguishable. A new release does not silently overwrite an older model or evaluation condition.