IROS 2026 in Pittsburgh runs September 27 through October 1

IEEE Robotics and Automation Society lists IROS 2026 in Pittsburgh from September 27 through October 1. Its venue record names the David L. Lawrence Convention Center. The IEEE RAS list of past and future IROS venues identifies Howie Choset as General Chair and Ross Hutton as Program Chair.

The official IEEE RAS event entry points readers to the conference website at 2026.ieee-iros.org. Program details can change during the event, so session times, room assignments and exhibitor information should be checked against the conference program rather than copied from an undated third-party list.

IEEE RAS published a set of community activities alongside the technical sessions and exhibits. On September 27 it scheduled the Conference Organizers Workshop. September 28 includes the Women in Engineering luncheon, the RAS Town Hall and a Student Activities Committee social hour.

On September 29, the published RAS schedule includes a Science Communication Crash Course, Young Professionals Luncheon, Student Activities Committee Lunch with Leaders and the RAS Standards Video Prize Competition dinner. These events sit beside the main conference program and give researchers routes into standards, mentoring and society governance.

Start with the physical platform. Record the robot model, end effector, sensors, compute hardware and control frequency. A manipulation result on a fixed arm does not establish the same performance on a mobile manipulator or humanoid.

Then record the evaluation conditions. Look for the number of trials, success definition, object set, initial-state variation, camera placement, operator intervention and reset policy. A percentage without these conditions is hard to compare with another paper.

For learning systems, separate pretraining, task-specific fine-tuning and test-time adaptation. If the method uses demonstrations, record how many, whether they came from the same robot and whether the test objects or scenes overlap with training.

A booth demo gives direct evidence that hardware can execute the shown sequence under the conditions in front of you. Ask whether the motion is teleoperated, scripted, replayed or generated by a policy. Record whether the operator selects actions, handles recovery or resets the scene.

Watch failure recovery as closely as successful motion. A robot that drops an object and recovers without a reset reveals something different from a selected successful take. For locomotion, note floor type, speed, turns, contact with people and whether external tracking or markers are used.

If a company publishes a payload, runtime or success figure, keep that number tied to the named robot version and test. Do not combine product specifications, research benchmarks and trade-show demonstrations into one maturity score.

The strongest material for follow-up coverage will include measured robot behavior, complete task definitions and primary sources. For humanoids and Physical AI, that means real-robot manipulation, locomotion under perturbation, tactile sensing, whole-body control, vision-language-action policies, teleoperation data and recovery after failure.

For each result, the article should preserve the boundary between simulation and hardware, autonomy and human control, one successful trial and repeated testing. That makes later comparisons easier when a similar system appears at another conference or moves into a customer pilot.