| Abstract [eng] |
Robotic systems have progressed from fenced industrial automation to co-located collaboration and, more recently, AI-mediated human–robot teaming. Haptic feedback provides a physical channel for intent communication, contact regulation, and safe interaction in close proximity. This structured narrative (scoping-style) mini-review focuses primarily on force-mediated and physically grounded haptic interaction – including impedance/admittance control, force exchange, contact regulation, and shared-control mechanisms – while discussing tactile, cutaneous, and vibrotactile modalities where they complement, localise, or communicate those interactions. It organises the field by addressing: (RQ1) how haptic modalities and devices evolve across interaction stages; (RQ2) which control and stability/safety strategies are used and what trade-offs they impose; and (RQ3) how performance is evaluated and what remains missing for reproducible comparison. Literature was identified through a structured scoping-style search and screening process across IEEE Xplore, Scopus, Web of Science, and Google Scholar (2000–2025); 63 key sources were retained for the final synthesis after duplicate removal, relevance screening, and backward/forward reference snowballing.For analytical clarity, we organise prior work into five interaction modes:caged robots (isolated industrial automation), human–robot interaction (HRI) including teleoperation and command-oriented haptic cues, shared-workspace human–robot collaboration (HRC), physical HRC (pHRC) with bidirectional force exchange, and AI-supported human–robot teaming (HRT). For each interaction mode, we distinguish between intentionally delivered haptic cues to the human (e.g., kinesthetic, tactile, cutaneous, or vibrotactile feedback), forces physically experienced during co-manipulation or incidental contact, and the robot-side sensing/compliance-control mechanisms (e.g., force/torque sensing and impedance/admittance) that shape or regulate those interactions, alongside the relevant safety constraints and commonly reported metrics (stability, transparency, workload, task efficiency, and safety). The main synthesis output is a design-space matrix that maps each interaction mode to the dominant haptic purpose, what the human mainly feels, robot-side enablers, primary design priorities, and typical evaluation focus/main risks. A single unified comparative modelling subsection, rather than four stage-specific derivations, supports this matrix by presenting the compact Equation (1)–Equation (11) progression only as an interpretive abstraction. Open problems include stable high-fidelity haptics under delay and uncertainty, reproducible benchmarking, and safety guidance for AI-adaptive haptic cues in teaming scenarios. |
