Robotics has entered an era of foundation models and data-driven policies, yet many benchmarks still emphasize utilitarian goals with binary outcomes. Embodied artistic creation offers a complementary frontier: robots must convert high-level intent into action while balancing hard constraints (contact, tool use, safety, deformable dynamics) with soft objectives (style, expressivity, aesthetics, human preference). Small modeling errors leave visible traces—“good” behavior is not only correct but also expressive, controllable, and interpretable.
Embodied4Arts reframes creative tasks as a driver for next-generation embodied intelligence: from generative models to generative motor programs. We aim to bring together robot learning, manipulation/control, world models, HRI, and creative AI/graphics to develop shared formulations and evaluation protocols.
Transform intent (text/image/sketch) into stroke programs under tool/material constraints. Focus: hierarchical action representations, verification/constraints, and metrics beyond pixel similarity.
Beyond note correctness: timing, dynamics, articulation, and interaction. Focus: contact/compliance, tempo/phrasing representations, and expressivity metrics.
Folding, weaving, sewing, knotting, draping: partial observability and complex state spaces. Focus: structured planning, learned/hybrid simulators, and standardized benchmarks.
Generate executable stroke trajectories in simulation to reproduce target sketches/paintings under a constrained budget (stroke count, motion cost, time).
Remote submission with organizer execution on a single standardized platform (same robot, tool mount, surface, and time budget). Teams submit executable programs/trajectories; organizers run all submissions under identical safety constraints and record outputs.
Sponsored by Axis.ai. Sponsorship funds will be used solely for prizes and workshop/community materials; it will not influence selection, review, or awards decisions, and no commercial promotion will be part of the technical program.
Status legend: Confirmed = speaker has explicitly agreed; Invited = pending response. We will verify that speakers/panelists accept a speaking/panel slot at at most one RSS workshop.
Panel format: invited speakers + organizers; we will add 1–2 panelists from RSS core communities to ensure strong alignment with planning/control/manipulation and evaluation.
Half-day (morning or afternoon). Includes a 30-minute coffee break that may overlap with a poster session.
| Time | Activity |
|---|---|
| 00:00–00:10 | Opening remarks: scope, goals, format, and challenge overview |
| 00:10–01:00 | Two invited talks (20–25 min + Q&A each): painting as programs; expressive music control |
| 01:00–01:30 | Coffee break + Poster session (posters remain up throughout) |
| 01:30–02:05 | Invited talk: deformables/craft |
| 02:05–02:30 | Contributed lightning talks (8–10 talks, 2–3 min each) |
| 02:30–02:50 | Challenge report + winner spotlights |
| 02:50–03:25 | Panel discussion + audience Q&A (moderated) |
| 03:25–03:55 | Breakout discussions + report-back (structured prompts) |
| 03:55–04:00 | Closing: outcomes, next steps, community actions |
Resources: standard A/V (projector, mics) and 20–40 poster boards; optional small tabletop demo area if space permits.
In-person: at least three organizers (including the primary contact) will attend.
Online (subject to RSS policy): we will post materials on the website and host an online Q&A channel.
Lightning talks will be selected from submissions (students/first-time presenters prioritized). Posters for accepted contributions.
We will publish the submission portal and detailed instructions on this page once the RSS workshop schedule is finalized.
Lead contact: Yanjia Huang — yanjia_0812@tamu.edu
Website: embodied4arts.github.io