Immersivecast opens imSim robot scoring portal, publishes 20 arm verdicts with failures

Immersivecast, a Seongnam robot simulation company founded in 2017, said on Oct 19 that it has opened imSim.app, a public portal where anyone can watch robots and AI models attempt standard tasks in physics simulation and see how each attempt was scored, without signing up. The company announced the launch in a statement carried by Money Today; the portal itself and the company's own site are the sources for every figure below. [Company profile](/companies/immersivecast).

The most unusual part is what the company publishes about itself. Its about page says it is not a certification or accreditation body, that every scoring run so far has been on its own simulations and robots with no customer-commissioned test yet, and that one comparison with a real robot on Sept 2 was judged insufficient evidence that simulated scores match reality.

Twenty arm verdicts, failures included

The first public table covers five robot arms on four standard tests, 20 verdicts measured on Oct 10 and 11 under scoring rules version 2.3b with imSim's default controller. Kinova's Gen3 placed parts most precisely at 1.48 mm, while the Franka Panda came in at 6.68 mm, about 4.5 times larger by our arithmetic. Across the five arms, 18 of 20 part-insertion trials succeeded (90%), with the UR10 and AgileX PiPER each missing one, and 119 of 120 measured self-locate pick-and-place attempts succeeded. The portal shows results it rates as caution or fail next to passes.

Beyond arms, the portal lists a delivery robot stopping at a parcel locker (8 of 8 checks passed, Oct 12), an urban drone patrol (7 of 7), a legged-robot terrain course (7 of 7, Oct 13) and a mass-scaling run in which the legged robot passed at its nominal mass and failed from 1.5 times, one of six stages. Users can upload a URDF file to get a joint summary, which the company says stays in the browser. Running tests on one's own AI opens first to invited members, with self-service robot registration to follow.

How the scoring works

imSim judges success from measured contact, position and force rather than a task log, uses three outcomes in which anything not measured is never counted as a pass, and reruns its scorer against tests that should fail to check that the scorer still discriminates. The method is published as an open specification, PAVS, first released in September and revised in October.