AWhat changed
Instead of naming a destination, a robot can be shown a photograph of the place it should reach. Zhu and colleagues combine that target image with the current camera view in a shared Siamese network, which processes both inputs using related parameters. An actor–critic policy then chooses movement commands. Sharing representations across destinations allows experience from one route to help another. The paper tests target changes, adaptation to new simulated rooms and a small real-room robot experiment. The distinctive contribution is the goal-conditioned navigation method; the simulator resource and inherited actor–critic and visual-recognition components retain their own credit.
Scientific scope
Instead of naming a destination, a robot can be shown a photograph of the place it should reach. Zhu and colleagues combine that target image with the current camera view in a shared Siamese network, which processes both inputs using related parameters. An actor–critic policy then chooses movement commands. Sharing representations across destinations allows experience from one route to help another. The paper tests target changes, adaptation to new simulated rooms and a small real-room robot experiment. The distinctive contribution is the goal-conditioned navigation method; the simulator resource and inherited actor–critic and visual-recognition components retain their own credit. The main simulated setup uses four movement commands, discretized positions and fixed-size turns. New destinations can still fail, especially far from trained targets. New rooms require training scene-specific layers; the real robot uses real images and fine-tuning. This is not universal zero-shot navigation, an optimal-path guarantee or a system learned without supervised visual pretraining. The small continuous-space example still uses the same four commands.
Evidence
The linked original works support the scoped statement and contribution attribution. WRS has not independently replayed experiments or reconstructed a full mathematical proof. The main simulated setup uses four movement commands, discretized positions and fixed-size turns. New destinations can still fail, especially far from trained targets. New rooms require training scene-specific layers; the real robot uses real images and fine-tuning. This is not universal zero-shot navigation, an optimal-path guarantee or a system learned without supervised visual pretraining. The small continuous-space example still uses the same four commands.
Contributors
- The named authors in the research record · SCIENCE: Source-listed human researchers and engineers for the exact method or resource. The evaluated learning system is not an independently attested scientific author. Earlier methods, funding and data-collection roles remain separate. Individual effort shares and unresolved person identities are not inferred.
Provisional WRS assessment, edition 2026-10-23.6. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction