Gravity cannot be switched off on a bench, so the researchers removed the directional information instead. A stem on a platform turning about its own axis still feels its weight, but the direction of "up" keeps moving, so there is no stable vector to grow away from; a surrounding sphere flooded with light arriving from all directions at once does the equivalent job on phototropism [2]. What remains for the tree to work with is its reading of its own shape [3].
The key observation concerns anatomy, not geometry. Once the device was running, the poplars stopped laying tension wood on the upper flank that had driven their initial upward bend, and began laying wood described as identical in every respect on the opposite face [7]. That symmetry is what makes the muscle analogy do work: the same contractile tissue that pulls a leaning stem up [12] can be deployed on the other side to pull an over-corrected stem back down. Until this work, tension wood was taken to form only on the upper side, the pattern you see at the base of trees on mountain slopes [8].
The sequence took a while. Trees spent roughly ten days bending upward before transfer [5], then straightened over the following weeks [6], which puts a single trial somewhere near three and a half to five and a half weeks per tree [16]. And because the sensory environment was manipulated rather than merely observed, the causal reading here is stronger than it would be from coring trees in a field survey: proprioception was the only cue left standing when the switch in tension wood placement occurred, which makes this a causal link rather than a coincidence of timing [3].
The thing this does not tell you is how much of it generalises. The INRAE announcement reports young poplar and no other species [15], gives no number of trees, no proportion that straightened and no statistical test [14], and describes no measurements on storm-damaged stands in the field [18]. Tension wood formation is a staged, cell-level process under proprioceptive control [13], which is a mechanism, not a dose-response curve.
INRAE frames the ability as contributing to tree resilience against storms and landslides, and says it matters more as the climate changes [11]. That is a reasonable reading of the mechanism, and it is also where the gap sits. A forester looking at two leaning stands after a gale now has a candidate explanation for divergent recovery, namely how well each stem senses its own curvature and how readily it commits tension wood to the corrective side, combined with its reading of light and gravity [17]. Nobody has yet measured that variation between genotypes, ages or sites. The same gap applies to the breeding pitch, where the group points to selecting cultivated plants on proprioception and to lodging in cereals [10]: a trait you cannot yet score cheaply in a nursery is not yet a selection criterion.
Worth noting what the 2012 result left open. A team involving INRAE showed then that plants have proprioception at all, without identifying the biological process it governs [9]. This paper, in New Phytologist [1], names the effector, which matters because an effector can be stained, timed and eventually bred for, in a way that an abstract sense cannot.
