EXOKERN

Force/torque-aware data and pretrained manipulation skills for contact-rich robotics. EXOKERN publishes datasets, pretrained policies, and evaluation tooling for assembly-style tasks where contact dynamics matter. The current public catalog focuses on peg insertion as a reproducible baseline for learning with 6-axis wrench signals, domain randomization, and multi-seed evaluation.

Start Here

Current Catalog

What Is Validated Today

• On the fixed-condition v0 benchmark, full_ft and no_ft both reached 100% success across 3 seeds x 100 episodes, while force/torque input reduced average contact force from 5.2 N to 3.2 N.
• On the domain-randomized v0.1.1 benchmark, both variants again reached 100% success across 600 total rollouts. In this setup, domain randomization largely closed the force gap, so the force/torque benefit is inconclusive for this specific task and tolerance regime.
• The cards state both the positive result and the boundary condition where a force/torque benefit does not appear. That is deliberate.

What Makes The Catalog Useful

• Every release is paired with an explicit dataset lineage, not a detached checkpoint drop.
• Baseline and ablation checkpoints are published together when possible.
• Evaluation claims are tied to concrete seeds, episode counts, and repository artifacts.
• Cards state simulation-only limitations directly instead of implying production readiness.
• The catalog is organized around contact-rich robotics rather than generic manipulation marketing.

Reference Papers Behind The Stack

• FORGE: Force-Guided Exploration for Robust Contact-Rich Manipulation under Uncertainty
• Factory: Factory: Fast Contact for Robotic Assembly
• IndustReal: Transferring Contact-Rich Assembly Tasks from Simulation to Reality
• Diffusion Policy: Visuomotor Policy Learning via Action Diffusion

Near-Term Priorities

• Harder contact tasks with tighter tolerances, threaded insertion, and snap-fit behavior
• Richer observations, including point clouds combined with wrench signals
• Multi-robot evaluation beyond a single arm configuration
• Real-world validation and safety-bridged deployment studies

External Links

• Website: exokern.com
• GitHub: github.com/Exokern
• X: x.com/ludwim_i
• Contact: exokern@proton.me EXOKERN is building a catalog for force-aware manipulation research: documented datasets, reproducible baselines, and claims that stay attached to the evidence.