Empirical studies
Research presenting experimental results on trustworthy embodied foundation models.
Security/CORE Unranked
RSS 2026
RSS 2026 Workshop on Trustworthy Embodied Foundation Models: Safety by Design versus Safety by Practice
RSS 2026 features a workshop on Trustworthy Embodied Foundation Models, focusing on safety in robotic systems powered by foundation models. The workshop explores two complementary approaches—safety by design (e.g., training data, architectures) and safety by practice (e.g., runtime monitoring, red-teaming)—to address risks like hallucination, poor grounding, and unsafe execution in real-world robotic applications.
Key deadlines
Verified September 9, 2026Abstract registration
June 21, 2026
AoE
Full paper
June 21, 2026
AoE
Workshop timeline
Submission and decisions
Abstract registrationKey deadline
June 21, 2026 · AoE
Full paperKey deadline
June 21, 2026 · AoE
Paper fit
Contribution paths
A strong submission should clearly identify its contribution and evaluate it appropriately.
Benchmarks
Datasets or evaluation frameworks for assessing safety in embodied foundation models.
Negative results
Studies reporting failed approaches or unexpected outcomes that inform future work.
Formal analyses
Theoretical or mathematical analyses of safety properties in foundation model-based robotics.
Deployment lessons
Practical insights from real-world deployment of foundation model-based robotic systems.
Tooling
Software or methodological tools that improve robustness, evaluation, or accountability of FM-based robotic autonomy.
Research areas in scope
01
Safety by Design
Continual learning for enhanced failure detection and recoveryRobust training regimens using real or simulated data (e.g., adversarial training, domain randomization)Benchmarking for embodied safety evaluationFormal methods and guaranteesStandardization for safety deployments in real-world and industrial applicationsExplainability for building trust in human–robot collaborations
02
Safety by Practice
Lessons from red-teaming and jailbreaking of VLA modelsRuntime assurance: monitoring, uncertainty/OOD detection, shielding, and safe fallback policiesHuman-in-the-loop systems for intervention, recovery, and incorporating failure feedback through intuitive interfaces
Policies worth checking twice
- Submissions are closed.
- Abstract and paper submission deadlines were specified and have passed.
- Camera-ready versions were due before the workshop date.
- Author notification occurred prior to the camera-ready deadline.
Official sources
Compiled from the official call for papers. The organizers’ pages remain authoritative.
Last verified September 9, 2026